]> git.karo-electronics.de Git - linux-beck.git/commitdiff
Orangefs: kernel client part 4
authorMike Marshall <hubcap@omnibond.com>
Fri, 17 Jul 2015 14:38:14 +0000 (10:38 -0400)
committerMike Marshall <hubcap@omnibond.com>
Sat, 3 Oct 2015 15:39:56 +0000 (11:39 -0400)
Signed-off-by: Mike Marshall <hubcap@omnibond.com>
fs/orangefs/pvfs2-sysfs.c [new file with mode: 0644]
fs/orangefs/pvfs2-utils.c [new file with mode: 0644]

diff --git a/fs/orangefs/pvfs2-sysfs.c b/fs/orangefs/pvfs2-sysfs.c
new file mode 100644 (file)
index 0000000..6d0e18b
--- /dev/null
@@ -0,0 +1,1787 @@
+/*
+ * Documentation/ABI/stable/orangefs-sysfs:
+ *
+ * What:               /sys/fs/orangefs/perf_counter_reset
+ * Date:               June 2015
+ * Contact:            Mike Marshall <hubcap@omnibond.com>
+ * Description:
+ *                     echo a 0 or a 1 into perf_counter_reset to
+ *                     reset all the counters in
+ *                     /sys/fs/orangefs/perf_counters
+ *                     except ones with PINT_PERF_PRESERVE set.
+ *
+ *
+ * What:               /sys/fs/orangefs/perf_counters/...
+ * Date:               Jun 2015
+ * Contact:            Mike Marshall <hubcap@omnibond.com>
+ * Description:
+ *                     Counters and settings for various caches.
+ *                     Read only.
+ *
+ *
+ * What:               /sys/fs/orangefs/perf_time_interval_secs
+ * Date:               Jun 2015
+ * Contact:            Mike Marshall <hubcap@omnibond.com>
+ * Description:
+ *                     Length of perf counter intervals in
+ *                     seconds.
+ *
+ *
+ * What:               /sys/fs/orangefs/perf_history_size
+ * Date:               Jun 2015
+ * Contact:            Mike Marshall <hubcap@omnibond.com>
+ * Description:
+ *                     The perf_counters cache statistics have N, or
+ *                     perf_history_size, samples. The default is
+ *                     one.
+ *
+ *                     Every perf_time_interval_secs the (first)
+ *                     samples are reset.
+ *
+ *                     If N is greater than one, the "current" set
+ *                     of samples is reset, and the samples from the
+ *                     other N-1 intervals remain available.
+ *
+ *
+ * What:               /sys/fs/orangefs/op_timeout_secs
+ * Date:               Jun 2015
+ * Contact:            Mike Marshall <hubcap@omnibond.com>
+ * Description:
+ *                     Service operation timeout in seconds.
+ *
+ *
+ * What:               /sys/fs/orangefs/slot_timeout_secs
+ * Date:               Jun 2015
+ * Contact:            Mike Marshall <hubcap@omnibond.com>
+ * Description:
+ *                     "Slot" timeout in seconds. A "slot"
+ *                     is an indexed buffer in the shared
+ *                     memory segment used for communication
+ *                     between the kernel module and userspace.
+ *                     Slots are requested and waited for,
+ *                     the wait times out after slot_timeout_secs.
+ *
+ *
+ * What:               /sys/fs/orangefs/acache/...
+ * Date:               Jun 2015
+ * Contact:            Mike Marshall <hubcap@omnibond.com>
+ * Description:
+ *                     Attribute cache configurable settings.
+ *
+ *
+ * What:               /sys/fs/orangefs/ncache/...
+ * Date:               Jun 2015
+ * Contact:            Mike Marshall <hubcap@omnibond.com>
+ * Description:
+ *                     Name cache configurable settings.
+ *
+ *
+ * What:               /sys/fs/orangefs/capcache/...
+ * Date:               Jun 2015
+ * Contact:            Mike Marshall <hubcap@omnibond.com>
+ * Description:
+ *                     Capability cache configurable settings.
+ *
+ *
+ * What:               /sys/fs/orangefs/ccache/...
+ * Date:               Jun 2015
+ * Contact:            Mike Marshall <hubcap@omnibond.com>
+ * Description:
+ *                     Credential cache configurable settings.
+ *
+ */
+
+#include <linux/fs.h>
+#include <linux/kobject.h>
+#include <linux/string.h>
+#include <linux/sysfs.h>
+#include <linux/module.h>
+#include <linux/init.h>
+
+#include "protocol.h"
+#include "pvfs2-kernel.h"
+#include "pvfs2-sysfs.h"
+
+#define ORANGEFS_KOBJ_ID "orangefs"
+#define ACACHE_KOBJ_ID "acache"
+#define CAPCACHE_KOBJ_ID "capcache"
+#define CCACHE_KOBJ_ID "ccache"
+#define NCACHE_KOBJ_ID "ncache"
+#define PC_KOBJ_ID "pc"
+#define STATS_KOBJ_ID "stats"
+
+struct orangefs_obj {
+       struct kobject kobj;
+       int op_timeout_secs;
+       int perf_counter_reset;
+       int perf_history_size;
+       int perf_time_interval_secs;
+       int slot_timeout_secs;
+};
+
+struct acache_orangefs_obj {
+       struct kobject kobj;
+       int hard_limit;
+       int reclaim_percentage;
+       int soft_limit;
+       int timeout_msecs;
+};
+
+struct capcache_orangefs_obj {
+       struct kobject kobj;
+       int hard_limit;
+       int reclaim_percentage;
+       int soft_limit;
+       int timeout_secs;
+};
+
+struct ccache_orangefs_obj {
+       struct kobject kobj;
+       int hard_limit;
+       int reclaim_percentage;
+       int soft_limit;
+       int timeout_secs;
+};
+
+struct ncache_orangefs_obj {
+       struct kobject kobj;
+       int hard_limit;
+       int reclaim_percentage;
+       int soft_limit;
+       int timeout_msecs;
+};
+
+struct pc_orangefs_obj {
+       struct kobject kobj;
+       char *acache;
+       char *capcache;
+       char *ncache;
+};
+
+struct stats_orangefs_obj {
+       struct kobject kobj;
+       int reads;
+       int writes;
+};
+
+struct orangefs_attribute {
+       struct attribute attr;
+       ssize_t (*show)(struct orangefs_obj *orangefs_obj,
+                       struct orangefs_attribute *attr,
+                       char *buf);
+       ssize_t (*store)(struct orangefs_obj *orangefs_obj,
+                        struct orangefs_attribute *attr,
+                        const char *buf,
+                        size_t count);
+};
+
+struct acache_orangefs_attribute {
+       struct attribute attr;
+       ssize_t (*show)(struct acache_orangefs_obj *acache_orangefs_obj,
+                       struct acache_orangefs_attribute *attr,
+                       char *buf);
+       ssize_t (*store)(struct acache_orangefs_obj *acache_orangefs_obj,
+                        struct acache_orangefs_attribute *attr,
+                        const char *buf,
+                        size_t count);
+};
+
+struct capcache_orangefs_attribute {
+       struct attribute attr;
+       ssize_t (*show)(struct capcache_orangefs_obj *capcache_orangefs_obj,
+                       struct capcache_orangefs_attribute *attr,
+                       char *buf);
+       ssize_t (*store)(struct capcache_orangefs_obj *capcache_orangefs_obj,
+                        struct capcache_orangefs_attribute *attr,
+                        const char *buf,
+                        size_t count);
+};
+
+struct ccache_orangefs_attribute {
+       struct attribute attr;
+       ssize_t (*show)(struct ccache_orangefs_obj *ccache_orangefs_obj,
+                       struct ccache_orangefs_attribute *attr,
+                       char *buf);
+       ssize_t (*store)(struct ccache_orangefs_obj *ccache_orangefs_obj,
+                        struct ccache_orangefs_attribute *attr,
+                        const char *buf,
+                        size_t count);
+};
+
+struct ncache_orangefs_attribute {
+       struct attribute attr;
+       ssize_t (*show)(struct ncache_orangefs_obj *ncache_orangefs_obj,
+                       struct ncache_orangefs_attribute *attr,
+                       char *buf);
+       ssize_t (*store)(struct ncache_orangefs_obj *ncache_orangefs_obj,
+                        struct ncache_orangefs_attribute *attr,
+                        const char *buf,
+                        size_t count);
+};
+
+struct pc_orangefs_attribute {
+       struct attribute attr;
+       ssize_t (*show)(struct pc_orangefs_obj *pc_orangefs_obj,
+                       struct pc_orangefs_attribute *attr,
+                       char *buf);
+       ssize_t (*store)(struct pc_orangefs_obj *pc_orangefs_obj,
+                        struct pc_orangefs_attribute *attr,
+                        const char *buf,
+                        size_t count);
+};
+
+struct stats_orangefs_attribute {
+       struct attribute attr;
+       ssize_t (*show)(struct stats_orangefs_obj *stats_orangefs_obj,
+                       struct stats_orangefs_attribute *attr,
+                       char *buf);
+       ssize_t (*store)(struct stats_orangefs_obj *stats_orangefs_obj,
+                        struct stats_orangefs_attribute *attr,
+                        const char *buf,
+                        size_t count);
+};
+
+static ssize_t orangefs_attr_show(struct kobject *kobj,
+                                 struct attribute *attr,
+                                 char *buf)
+{
+       struct orangefs_attribute *attribute;
+       struct orangefs_obj *orangefs_obj;
+       int rc;
+
+       attribute = container_of(attr, struct orangefs_attribute, attr);
+       orangefs_obj = container_of(kobj, struct orangefs_obj, kobj);
+
+       if (!attribute->show) {
+               rc = -EIO;
+               goto out;
+       }
+
+       rc = attribute->show(orangefs_obj, attribute, buf);
+
+out:
+       return rc;
+}
+
+static ssize_t orangefs_attr_store(struct kobject *kobj,
+                                  struct attribute *attr,
+                                  const char *buf,
+                                  size_t len)
+{
+       struct orangefs_attribute *attribute;
+       struct orangefs_obj *orangefs_obj;
+       int rc;
+
+       gossip_debug(GOSSIP_SYSFS_DEBUG,
+                    "orangefs_attr_store: start\n");
+
+       attribute = container_of(attr, struct orangefs_attribute, attr);
+       orangefs_obj = container_of(kobj, struct orangefs_obj, kobj);
+
+       if (!attribute->store) {
+               rc = -EIO;
+               goto out;
+       }
+
+       rc = attribute->store(orangefs_obj, attribute, buf, len);
+
+out:
+       return rc;
+}
+
+static const struct sysfs_ops orangefs_sysfs_ops = {
+       .show = orangefs_attr_show,
+       .store = orangefs_attr_store,
+};
+
+static ssize_t acache_orangefs_attr_show(struct kobject *kobj,
+                                        struct attribute *attr,
+                                        char *buf)
+{
+       struct acache_orangefs_attribute *attribute;
+       struct acache_orangefs_obj *acache_orangefs_obj;
+       int rc;
+
+       attribute = container_of(attr, struct acache_orangefs_attribute, attr);
+       acache_orangefs_obj =
+               container_of(kobj, struct acache_orangefs_obj, kobj);
+
+       if (!attribute->show) {
+               rc = -EIO;
+               goto out;
+       }
+
+       rc = attribute->show(acache_orangefs_obj, attribute, buf);
+
+out:
+       return rc;
+}
+
+static ssize_t acache_orangefs_attr_store(struct kobject *kobj,
+                                         struct attribute *attr,
+                                         const char *buf,
+                                         size_t len)
+{
+       struct acache_orangefs_attribute *attribute;
+       struct acache_orangefs_obj *acache_orangefs_obj;
+       int rc;
+
+       gossip_debug(GOSSIP_SYSFS_DEBUG,
+                    "acache_orangefs_attr_store: start\n");
+
+       attribute = container_of(attr, struct acache_orangefs_attribute, attr);
+       acache_orangefs_obj =
+               container_of(kobj, struct acache_orangefs_obj, kobj);
+
+       if (!attribute->store) {
+               rc = -EIO;
+               goto out;
+       }
+
+       rc = attribute->store(acache_orangefs_obj, attribute, buf, len);
+
+out:
+       return rc;
+}
+
+static const struct sysfs_ops acache_orangefs_sysfs_ops = {
+       .show = acache_orangefs_attr_show,
+       .store = acache_orangefs_attr_store,
+};
+
+static ssize_t capcache_orangefs_attr_show(struct kobject *kobj,
+                                          struct attribute *attr,
+                                          char *buf)
+{
+       struct capcache_orangefs_attribute *attribute;
+       struct capcache_orangefs_obj *capcache_orangefs_obj;
+       int rc;
+
+       attribute =
+               container_of(attr, struct capcache_orangefs_attribute, attr);
+       capcache_orangefs_obj =
+               container_of(kobj, struct capcache_orangefs_obj, kobj);
+
+       if (!attribute->show) {
+               rc = -EIO;
+               goto out;
+       }
+
+       rc = attribute->show(capcache_orangefs_obj, attribute, buf);
+
+out:
+       return rc;
+}
+
+static ssize_t capcache_orangefs_attr_store(struct kobject *kobj,
+                                           struct attribute *attr,
+                                           const char *buf,
+                                           size_t len)
+{
+       struct capcache_orangefs_attribute *attribute;
+       struct capcache_orangefs_obj *capcache_orangefs_obj;
+       int rc;
+
+       gossip_debug(GOSSIP_SYSFS_DEBUG,
+                    "capcache_orangefs_attr_store: start\n");
+
+       attribute =
+               container_of(attr, struct capcache_orangefs_attribute, attr);
+       capcache_orangefs_obj =
+               container_of(kobj, struct capcache_orangefs_obj, kobj);
+
+       if (!attribute->store) {
+               rc = -EIO;
+               goto out;
+       }
+
+       rc = attribute->store(capcache_orangefs_obj, attribute, buf, len);
+
+out:
+       return rc;
+}
+
+static const struct sysfs_ops capcache_orangefs_sysfs_ops = {
+       .show = capcache_orangefs_attr_show,
+       .store = capcache_orangefs_attr_store,
+};
+
+static ssize_t ccache_orangefs_attr_show(struct kobject *kobj,
+                                        struct attribute *attr,
+                                        char *buf)
+{
+       struct ccache_orangefs_attribute *attribute;
+       struct ccache_orangefs_obj *ccache_orangefs_obj;
+       int rc;
+
+       attribute =
+               container_of(attr, struct ccache_orangefs_attribute, attr);
+       ccache_orangefs_obj =
+               container_of(kobj, struct ccache_orangefs_obj, kobj);
+
+       if (!attribute->show) {
+               rc = -EIO;
+               goto out;
+       }
+
+       rc = attribute->show(ccache_orangefs_obj, attribute, buf);
+
+out:
+       return rc;
+}
+
+static ssize_t ccache_orangefs_attr_store(struct kobject *kobj,
+                                         struct attribute *attr,
+                                         const char *buf,
+                                         size_t len)
+{
+       struct ccache_orangefs_attribute *attribute;
+       struct ccache_orangefs_obj *ccache_orangefs_obj;
+       int rc;
+
+       gossip_debug(GOSSIP_SYSFS_DEBUG,
+                    "ccache_orangefs_attr_store: start\n");
+
+       attribute =
+               container_of(attr, struct ccache_orangefs_attribute, attr);
+       ccache_orangefs_obj =
+               container_of(kobj, struct ccache_orangefs_obj, kobj);
+
+       if (!attribute->store) {
+               rc = -EIO;
+               goto out;
+       }
+
+       rc = attribute->store(ccache_orangefs_obj, attribute, buf, len);
+
+out:
+       return rc;
+}
+
+static const struct sysfs_ops ccache_orangefs_sysfs_ops = {
+       .show = ccache_orangefs_attr_show,
+       .store = ccache_orangefs_attr_store,
+};
+
+static ssize_t ncache_orangefs_attr_show(struct kobject *kobj,
+                                        struct attribute *attr,
+                                        char *buf)
+{
+       struct ncache_orangefs_attribute *attribute;
+       struct ncache_orangefs_obj *ncache_orangefs_obj;
+       int rc;
+
+       attribute = container_of(attr, struct ncache_orangefs_attribute, attr);
+       ncache_orangefs_obj =
+               container_of(kobj, struct ncache_orangefs_obj, kobj);
+
+       if (!attribute->show) {
+               rc = -EIO;
+               goto out;
+       }
+
+       rc = attribute->show(ncache_orangefs_obj, attribute, buf);
+
+out:
+       return rc;
+}
+
+static ssize_t ncache_orangefs_attr_store(struct kobject *kobj,
+                                         struct attribute *attr,
+                                         const char *buf,
+                                         size_t len)
+{
+       struct ncache_orangefs_attribute *attribute;
+       struct ncache_orangefs_obj *ncache_orangefs_obj;
+       int rc;
+
+       gossip_debug(GOSSIP_SYSFS_DEBUG,
+                    "ncache_orangefs_attr_store: start\n");
+
+       attribute = container_of(attr, struct ncache_orangefs_attribute, attr);
+       ncache_orangefs_obj =
+               container_of(kobj, struct ncache_orangefs_obj, kobj);
+
+       if (!attribute->store) {
+               rc = -EIO;
+               goto out;
+       }
+
+       rc = attribute->store(ncache_orangefs_obj, attribute, buf, len);
+
+out:
+       return rc;
+}
+
+static const struct sysfs_ops ncache_orangefs_sysfs_ops = {
+       .show = ncache_orangefs_attr_show,
+       .store = ncache_orangefs_attr_store,
+};
+
+static ssize_t pc_orangefs_attr_show(struct kobject *kobj,
+                                    struct attribute *attr,
+                                    char *buf)
+{
+       struct pc_orangefs_attribute *attribute;
+       struct pc_orangefs_obj *pc_orangefs_obj;
+       int rc;
+
+       attribute = container_of(attr, struct pc_orangefs_attribute, attr);
+       pc_orangefs_obj =
+               container_of(kobj, struct pc_orangefs_obj, kobj);
+
+       if (!attribute->show) {
+               rc = -EIO;
+               goto out;
+       }
+
+       rc = attribute->show(pc_orangefs_obj, attribute, buf);
+
+out:
+       return rc;
+}
+
+static const struct sysfs_ops pc_orangefs_sysfs_ops = {
+       .show = pc_orangefs_attr_show,
+};
+
+static ssize_t stats_orangefs_attr_show(struct kobject *kobj,
+                                       struct attribute *attr,
+                                       char *buf)
+{
+       struct stats_orangefs_attribute *attribute;
+       struct stats_orangefs_obj *stats_orangefs_obj;
+       int rc;
+
+       attribute = container_of(attr, struct stats_orangefs_attribute, attr);
+       stats_orangefs_obj =
+               container_of(kobj, struct stats_orangefs_obj, kobj);
+
+       if (!attribute->show) {
+               rc = -EIO;
+               goto out;
+       }
+
+       rc = attribute->show(stats_orangefs_obj, attribute, buf);
+
+out:
+       return rc;
+}
+
+static const struct sysfs_ops stats_orangefs_sysfs_ops = {
+       .show = stats_orangefs_attr_show,
+};
+
+static void orangefs_release(struct kobject *kobj)
+{
+       struct orangefs_obj *orangefs_obj;
+
+       orangefs_obj = container_of(kobj, struct orangefs_obj, kobj);
+       kfree(orangefs_obj);
+}
+
+static void acache_orangefs_release(struct kobject *kobj)
+{
+       struct acache_orangefs_obj *acache_orangefs_obj;
+
+       acache_orangefs_obj =
+               container_of(kobj, struct acache_orangefs_obj, kobj);
+       kfree(acache_orangefs_obj);
+}
+
+static void capcache_orangefs_release(struct kobject *kobj)
+{
+       struct capcache_orangefs_obj *capcache_orangefs_obj;
+
+       capcache_orangefs_obj =
+               container_of(kobj, struct capcache_orangefs_obj, kobj);
+       kfree(capcache_orangefs_obj);
+}
+
+static void ccache_orangefs_release(struct kobject *kobj)
+{
+       struct ccache_orangefs_obj *ccache_orangefs_obj;
+
+       ccache_orangefs_obj =
+               container_of(kobj, struct ccache_orangefs_obj, kobj);
+       kfree(ccache_orangefs_obj);
+}
+
+static void ncache_orangefs_release(struct kobject *kobj)
+{
+       struct ncache_orangefs_obj *ncache_orangefs_obj;
+
+       ncache_orangefs_obj =
+               container_of(kobj, struct ncache_orangefs_obj, kobj);
+       kfree(ncache_orangefs_obj);
+}
+
+static void pc_orangefs_release(struct kobject *kobj)
+{
+       struct pc_orangefs_obj *pc_orangefs_obj;
+
+       pc_orangefs_obj =
+               container_of(kobj, struct pc_orangefs_obj, kobj);
+       kfree(pc_orangefs_obj);
+}
+
+static void stats_orangefs_release(struct kobject *kobj)
+{
+       struct stats_orangefs_obj *stats_orangefs_obj;
+
+       stats_orangefs_obj =
+               container_of(kobj, struct stats_orangefs_obj, kobj);
+       kfree(stats_orangefs_obj);
+}
+
+static ssize_t sysfs_int_show(char *kobj_id, char *buf, void *attr)
+{
+       int rc = -EIO;
+       struct orangefs_attribute *orangefs_attr;
+       struct stats_orangefs_attribute *stats_orangefs_attr;
+
+       gossip_debug(GOSSIP_SYSFS_DEBUG, "sysfs_int_show: id:%s:\n", kobj_id);
+
+       if (!strcmp(kobj_id, ORANGEFS_KOBJ_ID)) {
+               orangefs_attr = (struct orangefs_attribute *)attr;
+
+               if (!strcmp(orangefs_attr->attr.name, "op_timeout_secs")) {
+                       rc = scnprintf(buf,
+                                      PAGE_SIZE,
+                                      "%d\n",
+                                      op_timeout_secs);
+                       goto out;
+               } else if (!strcmp(orangefs_attr->attr.name,
+                                  "slot_timeout_secs")) {
+                       rc = scnprintf(buf,
+                                      PAGE_SIZE,
+                                      "%d\n",
+                                      slot_timeout_secs);
+                       goto out;
+               } else {
+                       goto out;
+               }
+
+       } else if (!strcmp(kobj_id, STATS_KOBJ_ID)) {
+               stats_orangefs_attr = (struct stats_orangefs_attribute *)attr;
+
+               if (!strcmp(stats_orangefs_attr->attr.name, "reads")) {
+                       rc = scnprintf(buf,
+                                      PAGE_SIZE,
+                                      "%lu\n",
+                                      g_pvfs2_stats.reads);
+                       goto out;
+               } else if (!strcmp(stats_orangefs_attr->attr.name, "writes")) {
+                       rc = scnprintf(buf,
+                                      PAGE_SIZE,
+                                      "%lu\n",
+                                      g_pvfs2_stats.writes);
+                       goto out;
+               } else {
+                       goto out;
+               }
+       }
+
+out:
+
+       return rc;
+}
+
+static ssize_t int_orangefs_show(struct orangefs_obj *orangefs_obj,
+                                struct orangefs_attribute *attr,
+                                char *buf)
+{
+       int rc;
+
+       gossip_debug(GOSSIP_SYSFS_DEBUG,
+                    "int_orangefs_show:start attr->attr.name:%s:\n",
+                    attr->attr.name);
+
+       rc = sysfs_int_show(ORANGEFS_KOBJ_ID, buf, (void *) attr);
+
+       return rc;
+}
+
+static ssize_t int_stats_show(struct stats_orangefs_obj *stats_orangefs_obj,
+                       struct stats_orangefs_attribute *attr,
+                       char *buf)
+{
+       int rc;
+
+       gossip_debug(GOSSIP_SYSFS_DEBUG,
+                    "int_stats_show:start attr->attr.name:%s:\n",
+                    attr->attr.name);
+
+       rc = sysfs_int_show(STATS_KOBJ_ID, buf, (void *) attr);
+
+       return rc;
+}
+
+static ssize_t int_store(struct orangefs_obj *orangefs_obj,
+                        struct orangefs_attribute *attr,
+                        const char *buf,
+                        size_t count)
+{
+       int rc = 0;
+
+       gossip_debug(GOSSIP_SYSFS_DEBUG,
+                    "int_store: start attr->attr.name:%s: buf:%s:\n",
+                    attr->attr.name, buf);
+
+       if (!strcmp(attr->attr.name, "op_timeout_secs")) {
+               rc = kstrtoint(buf, 0, &op_timeout_secs);
+               goto out;
+       } else if (!strcmp(attr->attr.name, "slot_timeout_secs")) {
+               rc = kstrtoint(buf, 0, &slot_timeout_secs);
+               goto out;
+       } else {
+               goto out;
+       }
+
+out:
+       if (rc)
+               rc = -EINVAL;
+       else
+               rc = count;
+
+       return rc;
+}
+
+/*
+ * obtain attribute values from userspace with a service operation.
+ */
+int sysfs_service_op_show(char *kobj_id, char *buf, void *attr)
+{
+       struct pvfs2_kernel_op_s *new_op = NULL;
+       int rc = 0;
+       char *ser_op_type = NULL;
+       struct orangefs_attribute *orangefs_attr;
+       struct acache_orangefs_attribute *acache_attr;
+       struct capcache_orangefs_attribute *capcache_attr;
+       struct ccache_orangefs_attribute *ccache_attr;
+       struct ncache_orangefs_attribute *ncache_attr;
+       struct pc_orangefs_attribute *pc_attr;
+       __u32 op_alloc_type;
+
+       gossip_debug(GOSSIP_SYSFS_DEBUG,
+                    "sysfs_service_op_show: id:%s:\n",
+                    kobj_id);
+
+       if (strcmp(kobj_id, PC_KOBJ_ID))
+               op_alloc_type = PVFS2_VFS_OP_PARAM;
+       else
+               op_alloc_type = PVFS2_VFS_OP_PERF_COUNT;
+
+       new_op = op_alloc(op_alloc_type);
+       if (!new_op) {
+               rc = -ENOMEM;
+               goto out;
+       }
+
+       /* Can't do a service_operation if the client is not running... */
+       rc = is_daemon_in_service();
+       if (rc) {
+               pr_info("%s: Client not running :%d:\n",
+                       __func__,
+                       is_daemon_in_service());
+               goto out;
+       }
+
+       if (strcmp(kobj_id, PC_KOBJ_ID))
+               new_op->upcall.req.param.type = PVFS2_PARAM_REQUEST_GET;
+
+       if (!strcmp(kobj_id, ORANGEFS_KOBJ_ID)) {
+               orangefs_attr = (struct orangefs_attribute *)attr;
+
+               if (!strcmp(orangefs_attr->attr.name, "perf_history_size"))
+                       new_op->upcall.req.param.op =
+                               PVFS2_PARAM_REQUEST_OP_PERF_HISTORY_SIZE;
+               else if (!strcmp(orangefs_attr->attr.name,
+                                "perf_time_interval_secs"))
+                       new_op->upcall.req.param.op =
+                               PVFS2_PARAM_REQUEST_OP_PERF_TIME_INTERVAL_SECS;
+               else if (!strcmp(orangefs_attr->attr.name,
+                                "perf_counter_reset"))
+                       new_op->upcall.req.param.op =
+                               PVFS2_PARAM_REQUEST_OP_PERF_RESET;
+
+       } else if (!strcmp(kobj_id, ACACHE_KOBJ_ID)) {
+               acache_attr = (struct acache_orangefs_attribute *)attr;
+
+               if (!strcmp(acache_attr->attr.name, "timeout_msecs"))
+                       new_op->upcall.req.param.op =
+                               PVFS2_PARAM_REQUEST_OP_ACACHE_TIMEOUT_MSECS;
+
+               if (!strcmp(acache_attr->attr.name, "hard_limit"))
+                       new_op->upcall.req.param.op =
+                               PVFS2_PARAM_REQUEST_OP_ACACHE_HARD_LIMIT;
+
+               if (!strcmp(acache_attr->attr.name, "soft_limit"))
+                       new_op->upcall.req.param.op =
+                               PVFS2_PARAM_REQUEST_OP_ACACHE_SOFT_LIMIT;
+
+               if (!strcmp(acache_attr->attr.name, "reclaim_percentage"))
+                       new_op->upcall.req.param.op =
+                         PVFS2_PARAM_REQUEST_OP_ACACHE_RECLAIM_PERCENTAGE;
+
+       } else if (!strcmp(kobj_id, CAPCACHE_KOBJ_ID)) {
+               capcache_attr = (struct capcache_orangefs_attribute *)attr;
+
+               if (!strcmp(capcache_attr->attr.name, "timeout_secs"))
+                       new_op->upcall.req.param.op =
+                               PVFS2_PARAM_REQUEST_OP_CAPCACHE_TIMEOUT_SECS;
+
+               if (!strcmp(capcache_attr->attr.name, "hard_limit"))
+                       new_op->upcall.req.param.op =
+                               PVFS2_PARAM_REQUEST_OP_CAPCACHE_HARD_LIMIT;
+
+               if (!strcmp(capcache_attr->attr.name, "soft_limit"))
+                       new_op->upcall.req.param.op =
+                               PVFS2_PARAM_REQUEST_OP_CAPCACHE_SOFT_LIMIT;
+
+               if (!strcmp(capcache_attr->attr.name, "reclaim_percentage"))
+                       new_op->upcall.req.param.op =
+                         PVFS2_PARAM_REQUEST_OP_CAPCACHE_RECLAIM_PERCENTAGE;
+
+       } else if (!strcmp(kobj_id, CCACHE_KOBJ_ID)) {
+               ccache_attr = (struct ccache_orangefs_attribute *)attr;
+
+               if (!strcmp(ccache_attr->attr.name, "timeout_secs"))
+                       new_op->upcall.req.param.op =
+                               PVFS2_PARAM_REQUEST_OP_CCACHE_TIMEOUT_SECS;
+
+               if (!strcmp(ccache_attr->attr.name, "hard_limit"))
+                       new_op->upcall.req.param.op =
+                               PVFS2_PARAM_REQUEST_OP_CCACHE_HARD_LIMIT;
+
+               if (!strcmp(ccache_attr->attr.name, "soft_limit"))
+                       new_op->upcall.req.param.op =
+                               PVFS2_PARAM_REQUEST_OP_CCACHE_SOFT_LIMIT;
+
+               if (!strcmp(ccache_attr->attr.name, "reclaim_percentage"))
+                       new_op->upcall.req.param.op =
+                         PVFS2_PARAM_REQUEST_OP_CCACHE_RECLAIM_PERCENTAGE;
+
+       } else if (!strcmp(kobj_id, NCACHE_KOBJ_ID)) {
+               ncache_attr = (struct ncache_orangefs_attribute *)attr;
+
+               if (!strcmp(ncache_attr->attr.name, "timeout_msecs"))
+                       new_op->upcall.req.param.op =
+                               PVFS2_PARAM_REQUEST_OP_NCACHE_TIMEOUT_MSECS;
+
+               if (!strcmp(ncache_attr->attr.name, "hard_limit"))
+                       new_op->upcall.req.param.op =
+                               PVFS2_PARAM_REQUEST_OP_NCACHE_HARD_LIMIT;
+
+               if (!strcmp(ncache_attr->attr.name, "soft_limit"))
+                       new_op->upcall.req.param.op =
+                               PVFS2_PARAM_REQUEST_OP_NCACHE_SOFT_LIMIT;
+
+               if (!strcmp(ncache_attr->attr.name, "reclaim_percentage"))
+                       new_op->upcall.req.param.op =
+                         PVFS2_PARAM_REQUEST_OP_NCACHE_RECLAIM_PERCENTAGE;
+
+       } else if (!strcmp(kobj_id, PC_KOBJ_ID)) {
+               pc_attr = (struct pc_orangefs_attribute *)attr;
+
+               if (!strcmp(pc_attr->attr.name, ACACHE_KOBJ_ID))
+                       new_op->upcall.req.perf_count.type =
+                               PVFS2_PERF_COUNT_REQUEST_ACACHE;
+
+               if (!strcmp(pc_attr->attr.name, CAPCACHE_KOBJ_ID))
+                       new_op->upcall.req.perf_count.type =
+                               PVFS2_PERF_COUNT_REQUEST_CAPCACHE;
+
+               if (!strcmp(pc_attr->attr.name, NCACHE_KOBJ_ID))
+                       new_op->upcall.req.perf_count.type =
+                               PVFS2_PERF_COUNT_REQUEST_NCACHE;
+
+       } else {
+               gossip_err("sysfs_service_op_show: unknown kobj_id:%s:\n",
+                          kobj_id);
+               rc = -EINVAL;
+               goto out;
+       }
+
+
+       if (strcmp(kobj_id, PC_KOBJ_ID))
+               ser_op_type = "pvfs2_param";
+       else
+               ser_op_type = "pvfs2_perf_count";
+
+       /*
+        * The service_operation will return an errno return code on
+        * error, and zero on success.
+        */
+       rc = service_operation(new_op, ser_op_type, PVFS2_OP_INTERRUPTIBLE);
+
+out:
+       if (!rc) {
+               if (strcmp(kobj_id, PC_KOBJ_ID)) {
+                       rc = scnprintf(buf,
+                                      PAGE_SIZE,
+                                      "%d\n",
+                                      (int)new_op->downcall.resp.param.value);
+               } else {
+                       rc = scnprintf(
+                               buf,
+                               PAGE_SIZE,
+                               "%s",
+                               new_op->downcall.resp.perf_count.buffer);
+               }
+       }
+
+       /*
+        * if we got ENOMEM, then op_alloc probably failed...
+        */
+       if (rc != -ENOMEM)
+               op_release(new_op);
+
+       return rc;
+
+}
+
+static ssize_t service_orangefs_show(struct orangefs_obj *orangefs_obj,
+                                    struct orangefs_attribute *attr,
+                                    char *buf)
+{
+       int rc = 0;
+
+       rc = sysfs_service_op_show(ORANGEFS_KOBJ_ID, buf, (void *)attr);
+
+       return rc;
+}
+
+static ssize_t
+       service_acache_show(struct acache_orangefs_obj *acache_orangefs_obj,
+                           struct acache_orangefs_attribute *attr,
+                           char *buf)
+{
+       int rc = 0;
+
+       rc = sysfs_service_op_show(ACACHE_KOBJ_ID, buf, (void *)attr);
+
+       return rc;
+}
+
+static ssize_t service_capcache_show(struct capcache_orangefs_obj
+                                       *capcache_orangefs_obj,
+                                    struct capcache_orangefs_attribute *attr,
+                                    char *buf)
+{
+       int rc = 0;
+
+       rc = sysfs_service_op_show(CAPCACHE_KOBJ_ID, buf, (void *)attr);
+
+       return rc;
+}
+
+static ssize_t service_ccache_show(struct ccache_orangefs_obj
+                                       *ccache_orangefs_obj,
+                                  struct ccache_orangefs_attribute *attr,
+                                  char *buf)
+{
+       int rc = 0;
+
+       rc = sysfs_service_op_show(CCACHE_KOBJ_ID, buf, (void *)attr);
+
+       return rc;
+}
+
+static ssize_t
+       service_ncache_show(struct ncache_orangefs_obj *ncache_orangefs_obj,
+                           struct ncache_orangefs_attribute *attr,
+                           char *buf)
+{
+       int rc = 0;
+
+       rc = sysfs_service_op_show(NCACHE_KOBJ_ID, buf, (void *)attr);
+
+       return rc;
+}
+
+static ssize_t
+       service_pc_show(struct pc_orangefs_obj *pc_orangefs_obj,
+                           struct pc_orangefs_attribute *attr,
+                           char *buf)
+{
+       int rc = 0;
+
+       rc = sysfs_service_op_show(PC_KOBJ_ID, buf, (void *)attr);
+
+       return rc;
+}
+
+/*
+ * pass attribute values back to userspace with a service operation.
+ *
+ * We have to do a memory allocation, an sscanf and a service operation.
+ * And we have to evaluate what the user entered, to make sure the
+ * value is within the range supported by the attribute. So, there's
+ * a lot of return code checking and mapping going on here.
+ *
+ * We want to return 1 if we think everything went OK, and
+ * EINVAL if not.
+ */
+int sysfs_service_op_store(char *kobj_id, const char *buf, void *attr)
+{
+       struct pvfs2_kernel_op_s *new_op = NULL;
+       int val = 0;
+       int rc = 0;
+       struct orangefs_attribute *orangefs_attr;
+       struct acache_orangefs_attribute *acache_attr;
+       struct capcache_orangefs_attribute *capcache_attr;
+       struct ccache_orangefs_attribute *ccache_attr;
+       struct ncache_orangefs_attribute *ncache_attr;
+
+       gossip_debug(GOSSIP_SYSFS_DEBUG,
+                    "sysfs_service_op_store: id:%s:\n",
+                    kobj_id);
+
+       new_op = op_alloc(PVFS2_VFS_OP_PARAM);
+       if (!new_op) {
+               rc = -ENOMEM;
+               goto out;
+       }
+
+       /* Can't do a service_operation if the client is not running... */
+       rc = is_daemon_in_service();
+       if (rc) {
+               pr_info("%s: Client not running :%d:\n",
+                       __func__,
+                       is_daemon_in_service());
+               goto out;
+       }
+
+       /*
+        * The value we want to send back to userspace is in buf.
+        */
+       rc = kstrtoint(buf, 0, &val);
+       if (rc)
+               goto out;
+
+       if (!strcmp(kobj_id, ORANGEFS_KOBJ_ID)) {
+               orangefs_attr = (struct orangefs_attribute *)attr;
+
+               if (!strcmp(orangefs_attr->attr.name, "perf_history_size")) {
+                       if (val > 0) {
+                               new_op->upcall.req.param.op =
+                                 PVFS2_PARAM_REQUEST_OP_PERF_HISTORY_SIZE;
+                       } else {
+                               rc = 0;
+                               goto out;
+                       }
+               } else if (!strcmp(orangefs_attr->attr.name,
+                                  "perf_time_interval_secs")) {
+                       if (val > 0) {
+                               new_op->upcall.req.param.op =
+                               PVFS2_PARAM_REQUEST_OP_PERF_TIME_INTERVAL_SECS;
+                       } else {
+                               rc = 0;
+                               goto out;
+                       }
+               } else if (!strcmp(orangefs_attr->attr.name,
+                                  "perf_counter_reset")) {
+                       if ((val == 0) || (val == 1)) {
+                               new_op->upcall.req.param.op =
+                                       PVFS2_PARAM_REQUEST_OP_PERF_RESET;
+                       } else {
+                               rc = 0;
+                               goto out;
+                       }
+               }
+
+       } else if (!strcmp(kobj_id, ACACHE_KOBJ_ID)) {
+               acache_attr = (struct acache_orangefs_attribute *)attr;
+
+               if (!strcmp(acache_attr->attr.name, "hard_limit")) {
+                       if (val > -1) {
+                               new_op->upcall.req.param.op =
+                                 PVFS2_PARAM_REQUEST_OP_ACACHE_HARD_LIMIT;
+                       } else {
+                               rc = 0;
+                               goto out;
+                       }
+               } else if (!strcmp(acache_attr->attr.name, "soft_limit")) {
+                       if (val > -1) {
+                               new_op->upcall.req.param.op =
+                                 PVFS2_PARAM_REQUEST_OP_ACACHE_SOFT_LIMIT;
+                       } else {
+                               rc = 0;
+                               goto out;
+                       }
+               } else if (!strcmp(acache_attr->attr.name,
+                                  "reclaim_percentage")) {
+                       if ((val > -1) && (val < 101)) {
+                               new_op->upcall.req.param.op =
+                                 PVFS2_PARAM_REQUEST_OP_ACACHE_RECLAIM_PERCENTAGE;
+                       } else {
+                               rc = 0;
+                               goto out;
+                       }
+               } else if (!strcmp(acache_attr->attr.name, "timeout_msecs")) {
+                       if (val > -1) {
+                               new_op->upcall.req.param.op =
+                                 PVFS2_PARAM_REQUEST_OP_ACACHE_TIMEOUT_MSECS;
+                       } else {
+                               rc = 0;
+                               goto out;
+                       }
+               }
+
+       } else if (!strcmp(kobj_id, CAPCACHE_KOBJ_ID)) {
+               capcache_attr = (struct capcache_orangefs_attribute *)attr;
+
+               if (!strcmp(capcache_attr->attr.name, "hard_limit")) {
+                       if (val > -1) {
+                               new_op->upcall.req.param.op =
+                                 PVFS2_PARAM_REQUEST_OP_CAPCACHE_HARD_LIMIT;
+                       } else {
+                               rc = 0;
+                               goto out;
+                       }
+               } else if (!strcmp(capcache_attr->attr.name, "soft_limit")) {
+                       if (val > -1) {
+                               new_op->upcall.req.param.op =
+                                 PVFS2_PARAM_REQUEST_OP_CAPCACHE_SOFT_LIMIT;
+                       } else {
+                               rc = 0;
+                               goto out;
+                       }
+               } else if (!strcmp(capcache_attr->attr.name,
+                                  "reclaim_percentage")) {
+                       if ((val > -1) && (val < 101)) {
+                               new_op->upcall.req.param.op =
+                                 PVFS2_PARAM_REQUEST_OP_CAPCACHE_RECLAIM_PERCENTAGE;
+                       } else {
+                               rc = 0;
+                               goto out;
+                       }
+               } else if (!strcmp(capcache_attr->attr.name, "timeout_secs")) {
+                       if (val > -1) {
+                               new_op->upcall.req.param.op =
+                                 PVFS2_PARAM_REQUEST_OP_CAPCACHE_TIMEOUT_SECS;
+                       } else {
+                               rc = 0;
+                               goto out;
+                       }
+               }
+
+       } else if (!strcmp(kobj_id, CCACHE_KOBJ_ID)) {
+               ccache_attr = (struct ccache_orangefs_attribute *)attr;
+
+               if (!strcmp(ccache_attr->attr.name, "hard_limit")) {
+                       if (val > -1) {
+                               new_op->upcall.req.param.op =
+                                 PVFS2_PARAM_REQUEST_OP_CCACHE_HARD_LIMIT;
+                       } else {
+                               rc = 0;
+                               goto out;
+                       }
+               } else if (!strcmp(ccache_attr->attr.name, "soft_limit")) {
+                       if (val > -1) {
+                               new_op->upcall.req.param.op =
+                                 PVFS2_PARAM_REQUEST_OP_CCACHE_SOFT_LIMIT;
+                       } else {
+                               rc = 0;
+                               goto out;
+                       }
+               } else if (!strcmp(ccache_attr->attr.name,
+                                  "reclaim_percentage")) {
+                       if ((val > -1) && (val < 101)) {
+                               new_op->upcall.req.param.op =
+                                 PVFS2_PARAM_REQUEST_OP_CCACHE_RECLAIM_PERCENTAGE;
+                       } else {
+                               rc = 0;
+                               goto out;
+                       }
+               } else if (!strcmp(ccache_attr->attr.name, "timeout_secs")) {
+                       if (val > -1) {
+                               new_op->upcall.req.param.op =
+                                 PVFS2_PARAM_REQUEST_OP_CCACHE_TIMEOUT_SECS;
+                       } else {
+                               rc = 0;
+                               goto out;
+                       }
+               }
+
+       } else if (!strcmp(kobj_id, NCACHE_KOBJ_ID)) {
+               ncache_attr = (struct ncache_orangefs_attribute *)attr;
+
+               if (!strcmp(ncache_attr->attr.name, "hard_limit")) {
+                       if (val > -1) {
+                               new_op->upcall.req.param.op =
+                                 PVFS2_PARAM_REQUEST_OP_NCACHE_HARD_LIMIT;
+                       } else {
+                               rc = 0;
+                               goto out;
+                       }
+               } else if (!strcmp(ncache_attr->attr.name, "soft_limit")) {
+                       if (val > -1) {
+                               new_op->upcall.req.param.op =
+                                 PVFS2_PARAM_REQUEST_OP_NCACHE_SOFT_LIMIT;
+                       } else {
+                               rc = 0;
+                               goto out;
+                       }
+               } else if (!strcmp(ncache_attr->attr.name,
+                                  "reclaim_percentage")) {
+                       if ((val > -1) && (val < 101)) {
+                               new_op->upcall.req.param.op =
+                                       PVFS2_PARAM_REQUEST_OP_NCACHE_RECLAIM_PERCENTAGE;
+                       } else {
+                               rc = 0;
+                               goto out;
+                       }
+               } else if (!strcmp(ncache_attr->attr.name, "timeout_msecs")) {
+                       if (val > -1) {
+                               new_op->upcall.req.param.op =
+                                 PVFS2_PARAM_REQUEST_OP_NCACHE_TIMEOUT_MSECS;
+                       } else {
+                               rc = 0;
+                               goto out;
+                       }
+               }
+
+       } else {
+               gossip_err("sysfs_service_op_store: unknown kobj_id:%s:\n",
+                          kobj_id);
+               rc = -EINVAL;
+               goto out;
+       }
+
+       new_op->upcall.req.param.type = PVFS2_PARAM_REQUEST_SET;
+
+       new_op->upcall.req.param.value = val;
+
+       /*
+        * The service_operation will return a errno return code on
+        * error, and zero on success.
+        */
+       rc = service_operation(new_op, "pvfs2_param", PVFS2_OP_INTERRUPTIBLE);
+
+       if (rc < 0) {
+               gossip_err("sysfs_service_op_store: service op returned:%d:\n",
+                       rc);
+               rc = 0;
+       } else {
+               rc = 1;
+       }
+
+out:
+       /*
+        * if we got ENOMEM, then op_alloc probably failed...
+        */
+       if (rc == -ENOMEM)
+               rc = 0;
+       else
+               op_release(new_op);
+
+       if (rc == 0)
+               rc = -EINVAL;
+
+       return rc;
+}
+
+static ssize_t
+       service_orangefs_store(struct orangefs_obj *orangefs_obj,
+                              struct orangefs_attribute *attr,
+                              const char *buf,
+                              size_t count)
+{
+       int rc = 0;
+
+       rc = sysfs_service_op_store(ORANGEFS_KOBJ_ID, buf, (void *) attr);
+
+       /* rc should have an errno value if the service_op went bad. */
+       if (rc == 1)
+               rc = count;
+
+       return rc;
+}
+
+static ssize_t
+       service_acache_store(struct acache_orangefs_obj *acache_orangefs_obj,
+                            struct acache_orangefs_attribute *attr,
+                            const char *buf,
+                            size_t count)
+{
+       int rc = 0;
+
+       rc = sysfs_service_op_store(ACACHE_KOBJ_ID, buf, (void *) attr);
+
+       /* rc should have an errno value if the service_op went bad. */
+       if (rc == 1)
+               rc = count;
+
+       return rc;
+}
+
+static ssize_t
+       service_capcache_store(struct capcache_orangefs_obj
+                               *capcache_orangefs_obj,
+                              struct capcache_orangefs_attribute *attr,
+                              const char *buf,
+                              size_t count)
+{
+       int rc = 0;
+
+       rc = sysfs_service_op_store(CAPCACHE_KOBJ_ID, buf, (void *) attr);
+
+       /* rc should have an errno value if the service_op went bad. */
+       if (rc == 1)
+               rc = count;
+
+       return rc;
+}
+
+static ssize_t service_ccache_store(struct ccache_orangefs_obj
+                                       *ccache_orangefs_obj,
+                                   struct ccache_orangefs_attribute *attr,
+                                   const char *buf,
+                                   size_t count)
+{
+       int rc = 0;
+
+       rc = sysfs_service_op_store(CCACHE_KOBJ_ID, buf, (void *) attr);
+
+       /* rc should have an errno value if the service_op went bad. */
+       if (rc == 1)
+               rc = count;
+
+       return rc;
+}
+
+static ssize_t
+       service_ncache_store(struct ncache_orangefs_obj *ncache_orangefs_obj,
+                            struct ncache_orangefs_attribute *attr,
+                            const char *buf,
+                            size_t count)
+{
+       int rc = 0;
+
+       rc = sysfs_service_op_store(NCACHE_KOBJ_ID, buf, (void *) attr);
+
+       /* rc should have an errno value if the service_op went bad. */
+       if (rc == 1)
+               rc = count;
+
+       return rc;
+}
+
+static struct orangefs_attribute op_timeout_secs_attribute =
+       __ATTR(op_timeout_secs, 0664, int_orangefs_show, int_store);
+
+static struct orangefs_attribute slot_timeout_secs_attribute =
+       __ATTR(slot_timeout_secs, 0664, int_orangefs_show, int_store);
+
+static struct orangefs_attribute perf_counter_reset_attribute =
+       __ATTR(perf_counter_reset,
+              0664,
+              service_orangefs_show,
+              service_orangefs_store);
+
+static struct orangefs_attribute perf_history_size_attribute =
+       __ATTR(perf_history_size,
+              0664,
+              service_orangefs_show,
+              service_orangefs_store);
+
+static struct orangefs_attribute perf_time_interval_secs_attribute =
+       __ATTR(perf_time_interval_secs,
+              0664,
+              service_orangefs_show,
+              service_orangefs_store);
+
+static struct attribute *orangefs_default_attrs[] = {
+       &op_timeout_secs_attribute.attr,
+       &slot_timeout_secs_attribute.attr,
+       &perf_counter_reset_attribute.attr,
+       &perf_history_size_attribute.attr,
+       &perf_time_interval_secs_attribute.attr,
+       NULL,
+};
+
+static struct kobj_type orangefs_ktype = {
+       .sysfs_ops = &orangefs_sysfs_ops,
+       .release = orangefs_release,
+       .default_attrs = orangefs_default_attrs,
+};
+
+static struct acache_orangefs_attribute acache_hard_limit_attribute =
+       __ATTR(hard_limit,
+              0664,
+              service_acache_show,
+              service_acache_store);
+
+static struct acache_orangefs_attribute acache_reclaim_percent_attribute =
+       __ATTR(reclaim_percentage,
+              0664,
+              service_acache_show,
+              service_acache_store);
+
+static struct acache_orangefs_attribute acache_soft_limit_attribute =
+       __ATTR(soft_limit,
+              0664,
+              service_acache_show,
+              service_acache_store);
+
+static struct acache_orangefs_attribute acache_timeout_msecs_attribute =
+       __ATTR(timeout_msecs,
+              0664,
+              service_acache_show,
+              service_acache_store);
+
+static struct attribute *acache_orangefs_default_attrs[] = {
+       &acache_hard_limit_attribute.attr,
+       &acache_reclaim_percent_attribute.attr,
+       &acache_soft_limit_attribute.attr,
+       &acache_timeout_msecs_attribute.attr,
+       NULL,
+};
+
+static struct kobj_type acache_orangefs_ktype = {
+       .sysfs_ops = &acache_orangefs_sysfs_ops,
+       .release = acache_orangefs_release,
+       .default_attrs = acache_orangefs_default_attrs,
+};
+
+static struct capcache_orangefs_attribute capcache_hard_limit_attribute =
+       __ATTR(hard_limit,
+              0664,
+              service_capcache_show,
+              service_capcache_store);
+
+static struct capcache_orangefs_attribute capcache_reclaim_percent_attribute =
+       __ATTR(reclaim_percentage,
+              0664,
+              service_capcache_show,
+              service_capcache_store);
+
+static struct capcache_orangefs_attribute capcache_soft_limit_attribute =
+       __ATTR(soft_limit,
+              0664,
+              service_capcache_show,
+              service_capcache_store);
+
+static struct capcache_orangefs_attribute capcache_timeout_secs_attribute =
+       __ATTR(timeout_secs,
+              0664,
+              service_capcache_show,
+              service_capcache_store);
+
+static struct attribute *capcache_orangefs_default_attrs[] = {
+       &capcache_hard_limit_attribute.attr,
+       &capcache_reclaim_percent_attribute.attr,
+       &capcache_soft_limit_attribute.attr,
+       &capcache_timeout_secs_attribute.attr,
+       NULL,
+};
+
+static struct kobj_type capcache_orangefs_ktype = {
+       .sysfs_ops = &capcache_orangefs_sysfs_ops,
+       .release = capcache_orangefs_release,
+       .default_attrs = capcache_orangefs_default_attrs,
+};
+
+static struct ccache_orangefs_attribute ccache_hard_limit_attribute =
+       __ATTR(hard_limit,
+              0664,
+              service_ccache_show,
+              service_ccache_store);
+
+static struct ccache_orangefs_attribute ccache_reclaim_percent_attribute =
+       __ATTR(reclaim_percentage,
+              0664,
+              service_ccache_show,
+              service_ccache_store);
+
+static struct ccache_orangefs_attribute ccache_soft_limit_attribute =
+       __ATTR(soft_limit,
+              0664,
+              service_ccache_show,
+              service_ccache_store);
+
+static struct ccache_orangefs_attribute ccache_timeout_secs_attribute =
+       __ATTR(timeout_secs,
+              0664,
+              service_ccache_show,
+              service_ccache_store);
+
+static struct attribute *ccache_orangefs_default_attrs[] = {
+       &ccache_hard_limit_attribute.attr,
+       &ccache_reclaim_percent_attribute.attr,
+       &ccache_soft_limit_attribute.attr,
+       &ccache_timeout_secs_attribute.attr,
+       NULL,
+};
+
+static struct kobj_type ccache_orangefs_ktype = {
+       .sysfs_ops = &ccache_orangefs_sysfs_ops,
+       .release = ccache_orangefs_release,
+       .default_attrs = ccache_orangefs_default_attrs,
+};
+
+static struct ncache_orangefs_attribute ncache_hard_limit_attribute =
+       __ATTR(hard_limit,
+              0664,
+              service_ncache_show,
+              service_ncache_store);
+
+static struct ncache_orangefs_attribute ncache_reclaim_percent_attribute =
+       __ATTR(reclaim_percentage,
+              0664,
+              service_ncache_show,
+              service_ncache_store);
+
+static struct ncache_orangefs_attribute ncache_soft_limit_attribute =
+       __ATTR(soft_limit,
+              0664,
+              service_ncache_show,
+              service_ncache_store);
+
+static struct ncache_orangefs_attribute ncache_timeout_msecs_attribute =
+       __ATTR(timeout_msecs,
+              0664,
+              service_ncache_show,
+              service_ncache_store);
+
+static struct attribute *ncache_orangefs_default_attrs[] = {
+       &ncache_hard_limit_attribute.attr,
+       &ncache_reclaim_percent_attribute.attr,
+       &ncache_soft_limit_attribute.attr,
+       &ncache_timeout_msecs_attribute.attr,
+       NULL,
+};
+
+static struct kobj_type ncache_orangefs_ktype = {
+       .sysfs_ops = &ncache_orangefs_sysfs_ops,
+       .release = ncache_orangefs_release,
+       .default_attrs = ncache_orangefs_default_attrs,
+};
+
+static struct pc_orangefs_attribute pc_acache_attribute =
+       __ATTR(acache,
+              0664,
+              service_pc_show,
+              NULL);
+
+static struct pc_orangefs_attribute pc_capcache_attribute =
+       __ATTR(capcache,
+              0664,
+              service_pc_show,
+              NULL);
+
+static struct pc_orangefs_attribute pc_ncache_attribute =
+       __ATTR(ncache,
+              0664,
+              service_pc_show,
+              NULL);
+
+static struct attribute *pc_orangefs_default_attrs[] = {
+       &pc_acache_attribute.attr,
+       &pc_capcache_attribute.attr,
+       &pc_ncache_attribute.attr,
+       NULL,
+};
+
+static struct kobj_type pc_orangefs_ktype = {
+       .sysfs_ops = &pc_orangefs_sysfs_ops,
+       .release = pc_orangefs_release,
+       .default_attrs = pc_orangefs_default_attrs,
+};
+
+static struct stats_orangefs_attribute stats_reads_attribute =
+       __ATTR(reads,
+              0664,
+              int_stats_show,
+              NULL);
+
+static struct stats_orangefs_attribute stats_writes_attribute =
+       __ATTR(writes,
+              0664,
+              int_stats_show,
+              NULL);
+
+static struct attribute *stats_orangefs_default_attrs[] = {
+       &stats_reads_attribute.attr,
+       &stats_writes_attribute.attr,
+       NULL,
+};
+
+static struct kobj_type stats_orangefs_ktype = {
+       .sysfs_ops = &stats_orangefs_sysfs_ops,
+       .release = stats_orangefs_release,
+       .default_attrs = stats_orangefs_default_attrs,
+};
+
+static struct orangefs_obj *orangefs_obj;
+static struct acache_orangefs_obj *acache_orangefs_obj;
+static struct capcache_orangefs_obj *capcache_orangefs_obj;
+static struct ccache_orangefs_obj *ccache_orangefs_obj;
+static struct ncache_orangefs_obj *ncache_orangefs_obj;
+static struct pc_orangefs_obj *pc_orangefs_obj;
+static struct stats_orangefs_obj *stats_orangefs_obj;
+
+int orangefs_sysfs_init(void)
+{
+       int rc;
+
+       gossip_debug(GOSSIP_SYSFS_DEBUG, "orangefs_sysfs_init: start\n");
+
+       /* create /sys/fs/orangefs. */
+       orangefs_obj = kzalloc(sizeof(*orangefs_obj), GFP_KERNEL);
+       if (!orangefs_obj) {
+               rc = -EINVAL;
+               goto out;
+       }
+
+       rc = kobject_init_and_add(&orangefs_obj->kobj,
+                                 &orangefs_ktype,
+                                 fs_kobj,
+                                 ORANGEFS_KOBJ_ID);
+
+       if (rc) {
+               kobject_put(&orangefs_obj->kobj);
+               rc = -EINVAL;
+               goto out;
+       }
+
+       kobject_uevent(&orangefs_obj->kobj, KOBJ_ADD);
+
+       /* create /sys/fs/orangefs/acache. */
+       acache_orangefs_obj = kzalloc(sizeof(*acache_orangefs_obj), GFP_KERNEL);
+       if (!acache_orangefs_obj) {
+               rc = -EINVAL;
+               goto out;
+       }
+
+       rc = kobject_init_and_add(&acache_orangefs_obj->kobj,
+                                 &acache_orangefs_ktype,
+                                 &orangefs_obj->kobj,
+                                 ACACHE_KOBJ_ID);
+
+       if (rc) {
+               kobject_put(&acache_orangefs_obj->kobj);
+               rc = -EINVAL;
+               goto out;
+       }
+
+       kobject_uevent(&acache_orangefs_obj->kobj, KOBJ_ADD);
+
+       /* create /sys/fs/orangefs/capcache. */
+       capcache_orangefs_obj =
+               kzalloc(sizeof(*capcache_orangefs_obj), GFP_KERNEL);
+       if (!capcache_orangefs_obj) {
+               rc = -EINVAL;
+               goto out;
+       }
+
+       rc = kobject_init_and_add(&capcache_orangefs_obj->kobj,
+                                 &capcache_orangefs_ktype,
+                                 &orangefs_obj->kobj,
+                                 CAPCACHE_KOBJ_ID);
+       if (rc) {
+               kobject_put(&capcache_orangefs_obj->kobj);
+               rc = -EINVAL;
+               goto out;
+       }
+
+       kobject_uevent(&capcache_orangefs_obj->kobj, KOBJ_ADD);
+
+       /* create /sys/fs/orangefs/ccache. */
+       ccache_orangefs_obj =
+               kzalloc(sizeof(*ccache_orangefs_obj), GFP_KERNEL);
+       if (!ccache_orangefs_obj) {
+               rc = -EINVAL;
+               goto out;
+       }
+
+       rc = kobject_init_and_add(&ccache_orangefs_obj->kobj,
+                                 &ccache_orangefs_ktype,
+                                 &orangefs_obj->kobj,
+                                 CCACHE_KOBJ_ID);
+       if (rc) {
+               kobject_put(&ccache_orangefs_obj->kobj);
+               rc = -EINVAL;
+               goto out;
+       }
+
+       kobject_uevent(&ccache_orangefs_obj->kobj, KOBJ_ADD);
+
+       /* create /sys/fs/orangefs/ncache. */
+       ncache_orangefs_obj = kzalloc(sizeof(*ncache_orangefs_obj), GFP_KERNEL);
+       if (!ncache_orangefs_obj) {
+               rc = -EINVAL;
+               goto out;
+       }
+
+       rc = kobject_init_and_add(&ncache_orangefs_obj->kobj,
+                                 &ncache_orangefs_ktype,
+                                 &orangefs_obj->kobj,
+                                 NCACHE_KOBJ_ID);
+
+       if (rc) {
+               kobject_put(&ncache_orangefs_obj->kobj);
+               rc = -EINVAL;
+               goto out;
+       }
+
+       kobject_uevent(&ncache_orangefs_obj->kobj, KOBJ_ADD);
+
+       /* create /sys/fs/orangefs/perf_counters. */
+       pc_orangefs_obj = kzalloc(sizeof(*pc_orangefs_obj), GFP_KERNEL);
+       if (!pc_orangefs_obj) {
+               rc = -EINVAL;
+               goto out;
+       }
+
+       rc = kobject_init_and_add(&pc_orangefs_obj->kobj,
+                                 &pc_orangefs_ktype,
+                                 &orangefs_obj->kobj,
+                                 "perf_counters");
+
+       if (rc) {
+               kobject_put(&pc_orangefs_obj->kobj);
+               rc = -EINVAL;
+               goto out;
+       }
+
+       kobject_uevent(&pc_orangefs_obj->kobj, KOBJ_ADD);
+
+       /* create /sys/fs/orangefs/stats. */
+       stats_orangefs_obj = kzalloc(sizeof(*stats_orangefs_obj), GFP_KERNEL);
+       if (!stats_orangefs_obj) {
+               rc = -EINVAL;
+               goto out;
+       }
+
+       rc = kobject_init_and_add(&stats_orangefs_obj->kobj,
+                                 &stats_orangefs_ktype,
+                                 &orangefs_obj->kobj,
+                                 STATS_KOBJ_ID);
+
+       if (rc) {
+               kobject_put(&stats_orangefs_obj->kobj);
+               rc = -EINVAL;
+               goto out;
+       }
+
+       kobject_uevent(&stats_orangefs_obj->kobj, KOBJ_ADD);
+out:
+       return rc;
+}
+
+void orangefs_sysfs_exit(void)
+{
+       gossip_debug(GOSSIP_SYSFS_DEBUG, "orangefs_sysfs_exit: start\n");
+
+       kobject_put(&acache_orangefs_obj->kobj);
+       kobject_put(&capcache_orangefs_obj->kobj);
+       kobject_put(&ccache_orangefs_obj->kobj);
+       kobject_put(&ncache_orangefs_obj->kobj);
+       kobject_put(&pc_orangefs_obj->kobj);
+       kobject_put(&stats_orangefs_obj->kobj);
+
+       kobject_put(&orangefs_obj->kobj);
+}
diff --git a/fs/orangefs/pvfs2-utils.c b/fs/orangefs/pvfs2-utils.c
new file mode 100644 (file)
index 0000000..107f425
--- /dev/null
@@ -0,0 +1,1128 @@
+/*
+ * (C) 2001 Clemson University and The University of Chicago
+ *
+ * See COPYING in top-level directory.
+ */
+#include "protocol.h"
+#include "pvfs2-kernel.h"
+#include "pvfs2-dev-proto.h"
+#include "pvfs2-bufmap.h"
+
+__s32 fsid_of_op(struct pvfs2_kernel_op_s *op)
+{
+       __s32 fsid = PVFS_FS_ID_NULL;
+
+       if (op) {
+               switch (op->upcall.type) {
+               case PVFS2_VFS_OP_FILE_IO:
+                       fsid = op->upcall.req.io.refn.fs_id;
+                       break;
+               case PVFS2_VFS_OP_LOOKUP:
+                       fsid = op->upcall.req.lookup.parent_refn.fs_id;
+                       break;
+               case PVFS2_VFS_OP_CREATE:
+                       fsid = op->upcall.req.create.parent_refn.fs_id;
+                       break;
+               case PVFS2_VFS_OP_GETATTR:
+                       fsid = op->upcall.req.getattr.refn.fs_id;
+                       break;
+               case PVFS2_VFS_OP_REMOVE:
+                       fsid = op->upcall.req.remove.parent_refn.fs_id;
+                       break;
+               case PVFS2_VFS_OP_MKDIR:
+                       fsid = op->upcall.req.mkdir.parent_refn.fs_id;
+                       break;
+               case PVFS2_VFS_OP_READDIR:
+                       fsid = op->upcall.req.readdir.refn.fs_id;
+                       break;
+               case PVFS2_VFS_OP_SETATTR:
+                       fsid = op->upcall.req.setattr.refn.fs_id;
+                       break;
+               case PVFS2_VFS_OP_SYMLINK:
+                       fsid = op->upcall.req.sym.parent_refn.fs_id;
+                       break;
+               case PVFS2_VFS_OP_RENAME:
+                       fsid = op->upcall.req.rename.old_parent_refn.fs_id;
+                       break;
+               case PVFS2_VFS_OP_STATFS:
+                       fsid = op->upcall.req.statfs.fs_id;
+                       break;
+               case PVFS2_VFS_OP_TRUNCATE:
+                       fsid = op->upcall.req.truncate.refn.fs_id;
+                       break;
+               case PVFS2_VFS_OP_MMAP_RA_FLUSH:
+                       fsid = op->upcall.req.ra_cache_flush.refn.fs_id;
+                       break;
+               case PVFS2_VFS_OP_FS_UMOUNT:
+                       fsid = op->upcall.req.fs_umount.fs_id;
+                       break;
+               case PVFS2_VFS_OP_GETXATTR:
+                       fsid = op->upcall.req.getxattr.refn.fs_id;
+                       break;
+               case PVFS2_VFS_OP_SETXATTR:
+                       fsid = op->upcall.req.setxattr.refn.fs_id;
+                       break;
+               case PVFS2_VFS_OP_LISTXATTR:
+                       fsid = op->upcall.req.listxattr.refn.fs_id;
+                       break;
+               case PVFS2_VFS_OP_REMOVEXATTR:
+                       fsid = op->upcall.req.removexattr.refn.fs_id;
+                       break;
+               case PVFS2_VFS_OP_FSYNC:
+                       fsid = op->upcall.req.fsync.refn.fs_id;
+                       break;
+               default:
+                       break;
+               }
+       }
+       return fsid;
+}
+
+static void pvfs2_set_inode_flags(struct inode *inode,
+                                 struct PVFS_sys_attr_s *attrs)
+{
+       if (attrs->flags & PVFS_IMMUTABLE_FL)
+               inode->i_flags |= S_IMMUTABLE;
+       else
+               inode->i_flags &= ~S_IMMUTABLE;
+
+       if (attrs->flags & PVFS_APPEND_FL)
+               inode->i_flags |= S_APPEND;
+       else
+               inode->i_flags &= ~S_APPEND;
+
+       if (attrs->flags & PVFS_NOATIME_FL)
+               inode->i_flags |= S_NOATIME;
+       else
+               inode->i_flags &= ~S_NOATIME;
+
+}
+
+/* NOTE: symname is ignored unless the inode is a sym link */
+static int copy_attributes_to_inode(struct inode *inode,
+                                   struct PVFS_sys_attr_s *attrs,
+                                   char *symname)
+{
+       int ret = -1;
+       int perm_mode = 0;
+       struct pvfs2_inode_s *pvfs2_inode = PVFS2_I(inode);
+       loff_t inode_size = 0;
+       loff_t rounded_up_size = 0;
+
+
+       /*
+          arbitrarily set the inode block size; FIXME: we need to
+          resolve the difference between the reported inode blocksize
+          and the PAGE_CACHE_SIZE, since our block count will always
+          be wrong.
+
+          For now, we're setting the block count to be the proper
+          number assuming the block size is 512 bytes, and the size is
+          rounded up to the nearest 4K.  This is apparently required
+          to get proper size reports from the 'du' shell utility.
+
+          changing the inode->i_blkbits to something other than
+          PAGE_CACHE_SHIFT breaks mmap/execution as we depend on that.
+        */
+       gossip_debug(GOSSIP_UTILS_DEBUG,
+                    "attrs->mask = %x (objtype = %s)\n",
+                    attrs->mask,
+                    attrs->objtype == PVFS_TYPE_METAFILE ? "file" :
+                    attrs->objtype == PVFS_TYPE_DIRECTORY ? "directory" :
+                    attrs->objtype == PVFS_TYPE_SYMLINK ? "symlink" :
+                       "invalid/unknown");
+
+       switch (attrs->objtype) {
+       case PVFS_TYPE_METAFILE:
+               pvfs2_set_inode_flags(inode, attrs);
+               if (attrs->mask & PVFS_ATTR_SYS_SIZE) {
+                       inode_size = (loff_t) attrs->size;
+                       rounded_up_size =
+                           (inode_size + (4096 - (inode_size % 4096)));
+
+                       pvfs2_lock_inode(inode);
+                       inode->i_bytes = inode_size;
+                       inode->i_blocks =
+                           (unsigned long)(rounded_up_size / 512);
+                       pvfs2_unlock_inode(inode);
+
+                       /*
+                        * NOTE: make sure all the places we're called
+                        * from have the inode->i_sem lock. We're fine
+                        * in 99% of the cases since we're mostly
+                        * called from a lookup.
+                        */
+                       inode->i_size = inode_size;
+               }
+               break;
+       case PVFS_TYPE_SYMLINK:
+               if (symname != NULL) {
+                       inode->i_size = (loff_t) strlen(symname);
+                       break;
+               }
+               /*FALLTHRU*/
+       default:
+               pvfs2_lock_inode(inode);
+               inode->i_bytes = PAGE_CACHE_SIZE;
+               inode->i_blocks = (unsigned long)(PAGE_CACHE_SIZE / 512);
+               pvfs2_unlock_inode(inode);
+
+               inode->i_size = PAGE_CACHE_SIZE;
+               break;
+       }
+
+       inode->i_uid = make_kuid(&init_user_ns, attrs->owner);
+       inode->i_gid = make_kgid(&init_user_ns, attrs->group);
+       inode->i_atime.tv_sec = (time_t) attrs->atime;
+       inode->i_mtime.tv_sec = (time_t) attrs->mtime;
+       inode->i_ctime.tv_sec = (time_t) attrs->ctime;
+       inode->i_atime.tv_nsec = 0;
+       inode->i_mtime.tv_nsec = 0;
+       inode->i_ctime.tv_nsec = 0;
+
+       if (attrs->perms & PVFS_O_EXECUTE)
+               perm_mode |= S_IXOTH;
+       if (attrs->perms & PVFS_O_WRITE)
+               perm_mode |= S_IWOTH;
+       if (attrs->perms & PVFS_O_READ)
+               perm_mode |= S_IROTH;
+
+       if (attrs->perms & PVFS_G_EXECUTE)
+               perm_mode |= S_IXGRP;
+       if (attrs->perms & PVFS_G_WRITE)
+               perm_mode |= S_IWGRP;
+       if (attrs->perms & PVFS_G_READ)
+               perm_mode |= S_IRGRP;
+
+       if (attrs->perms & PVFS_U_EXECUTE)
+               perm_mode |= S_IXUSR;
+       if (attrs->perms & PVFS_U_WRITE)
+               perm_mode |= S_IWUSR;
+       if (attrs->perms & PVFS_U_READ)
+               perm_mode |= S_IRUSR;
+
+       if (attrs->perms & PVFS_G_SGID)
+               perm_mode |= S_ISGID;
+       if (attrs->perms & PVFS_U_SUID)
+               perm_mode |= S_ISUID;
+
+       inode->i_mode = perm_mode;
+
+       if (is_root_handle(inode)) {
+               /* special case: mark the root inode as sticky */
+               inode->i_mode |= S_ISVTX;
+               gossip_debug(GOSSIP_UTILS_DEBUG,
+                            "Marking inode %pU as sticky\n",
+                            get_khandle_from_ino(inode));
+       }
+
+       switch (attrs->objtype) {
+       case PVFS_TYPE_METAFILE:
+               inode->i_mode |= S_IFREG;
+               ret = 0;
+               break;
+       case PVFS_TYPE_DIRECTORY:
+               inode->i_mode |= S_IFDIR;
+               /* NOTE: we have no good way to keep nlink consistent
+                * for directories across clients; keep constant at 1.
+                * Why 1?  If we go with 2, then find(1) gets confused
+                * and won't work properly withouth the -noleaf option
+                */
+               set_nlink(inode, 1);
+               ret = 0;
+               break;
+       case PVFS_TYPE_SYMLINK:
+               inode->i_mode |= S_IFLNK;
+
+               /* copy link target to inode private data */
+               if (pvfs2_inode && symname) {
+                       strncpy(pvfs2_inode->link_target,
+                               symname,
+                               PVFS_NAME_MAX);
+                       gossip_debug(GOSSIP_UTILS_DEBUG,
+                                    "Copied attr link target %s\n",
+                                    pvfs2_inode->link_target);
+               }
+               gossip_debug(GOSSIP_UTILS_DEBUG,
+                            "symlink mode %o\n",
+                            inode->i_mode);
+               ret = 0;
+               break;
+       default:
+               gossip_err("pvfs2: copy_attributes_to_inode: got invalid attribute type %x\n",
+                       attrs->objtype);
+       }
+
+       gossip_debug(GOSSIP_UTILS_DEBUG,
+                    "pvfs2: copy_attributes_to_inode: setting i_mode to %o, i_size to %lu\n",
+                    inode->i_mode,
+                    (unsigned long)i_size_read(inode));
+
+       return ret;
+}
+
+/*
+ * NOTE: in kernel land, we never use the sys_attr->link_target for
+ * anything, so don't bother copying it into the sys_attr object here.
+ */
+static inline int copy_attributes_from_inode(struct inode *inode,
+                                            struct PVFS_sys_attr_s *attrs,
+                                            struct iattr *iattr)
+{
+       umode_t tmp_mode;
+
+       if (!iattr || !inode || !attrs) {
+               gossip_err("NULL iattr (%p), inode (%p), attrs (%p) "
+                          "in copy_attributes_from_inode!\n",
+                          iattr,
+                          inode,
+                          attrs);
+               return -EINVAL;
+       }
+       /*
+        * We need to be careful to only copy the attributes out of the
+        * iattr object that we know are valid.
+        */
+       attrs->mask = 0;
+       if (iattr->ia_valid & ATTR_UID) {
+               attrs->owner = from_kuid(current_user_ns(), iattr->ia_uid);
+               attrs->mask |= PVFS_ATTR_SYS_UID;
+               gossip_debug(GOSSIP_UTILS_DEBUG, "(UID) %d\n", attrs->owner);
+       }
+       if (iattr->ia_valid & ATTR_GID) {
+               attrs->group = from_kgid(current_user_ns(), iattr->ia_gid);
+               attrs->mask |= PVFS_ATTR_SYS_GID;
+               gossip_debug(GOSSIP_UTILS_DEBUG, "(GID) %d\n", attrs->group);
+       }
+
+       if (iattr->ia_valid & ATTR_ATIME) {
+               attrs->mask |= PVFS_ATTR_SYS_ATIME;
+               if (iattr->ia_valid & ATTR_ATIME_SET) {
+                       attrs->atime =
+                           pvfs2_convert_time_field((void *)&iattr->ia_atime);
+                       attrs->mask |= PVFS_ATTR_SYS_ATIME_SET;
+               }
+       }
+       if (iattr->ia_valid & ATTR_MTIME) {
+               attrs->mask |= PVFS_ATTR_SYS_MTIME;
+               if (iattr->ia_valid & ATTR_MTIME_SET) {
+                       attrs->mtime =
+                           pvfs2_convert_time_field((void *)&iattr->ia_mtime);
+                       attrs->mask |= PVFS_ATTR_SYS_MTIME_SET;
+               }
+       }
+       if (iattr->ia_valid & ATTR_CTIME)
+               attrs->mask |= PVFS_ATTR_SYS_CTIME;
+
+       /*
+        * PVFS2 cannot set size with a setattr operation.  Probably not likely
+        * to be requested through the VFS, but just in case, don't worry about
+        * ATTR_SIZE
+        */
+
+       if (iattr->ia_valid & ATTR_MODE) {
+               tmp_mode = iattr->ia_mode;
+               if (tmp_mode & (S_ISVTX)) {
+                       if (is_root_handle(inode)) {
+                               /*
+                                * allow sticky bit to be set on root (since
+                                * it shows up that way by default anyhow),
+                                * but don't show it to the server
+                                */
+                               tmp_mode -= S_ISVTX;
+                       } else {
+                               gossip_debug(GOSSIP_UTILS_DEBUG,
+                                            "User attempted to set sticky bit on non-root directory; returning EINVAL.\n");
+                               return -EINVAL;
+                       }
+               }
+
+               if (tmp_mode & (S_ISUID)) {
+                       gossip_debug(GOSSIP_UTILS_DEBUG,
+                                    "Attempting to set setuid bit (not supported); returning EINVAL.\n");
+                       return -EINVAL;
+               }
+
+               attrs->perms = PVFS_util_translate_mode(tmp_mode);
+               attrs->mask |= PVFS_ATTR_SYS_PERM;
+       }
+
+       return 0;
+}
+
+/*
+ * issues a pvfs2 getattr request and fills in the appropriate inode
+ * attributes if successful.  returns 0 on success; -errno otherwise
+ */
+int pvfs2_inode_getattr(struct inode *inode, __u32 getattr_mask)
+{
+       struct pvfs2_inode_s *pvfs2_inode = PVFS2_I(inode);
+       struct pvfs2_kernel_op_s *new_op;
+       int ret = -EINVAL;
+
+       gossip_debug(GOSSIP_UTILS_DEBUG,
+                    "%s: called on inode %pU\n",
+                    __func__,
+                    get_khandle_from_ino(inode));
+
+       new_op = op_alloc(PVFS2_VFS_OP_GETATTR);
+       if (!new_op)
+               return -ENOMEM;
+       new_op->upcall.req.getattr.refn = pvfs2_inode->refn;
+       new_op->upcall.req.getattr.mask = getattr_mask;
+
+       ret = service_operation(new_op, __func__,
+                               get_interruptible_flag(inode));
+       if (ret != 0)
+               goto out;
+
+       if (copy_attributes_to_inode(inode,
+                       &new_op->downcall.resp.getattr.attributes,
+                       new_op->downcall.resp.getattr.link_target)) {
+               gossip_err("%s: failed to copy attributes\n", __func__);
+               ret = -ENOENT;
+               goto out;
+       }
+
+       /*
+        * Store blksize in pvfs2 specific part of inode structure; we are
+        * only going to use this to report to stat to make sure it doesn't
+        * perturb any inode related code paths.
+        */
+       if (new_op->downcall.resp.getattr.attributes.objtype ==
+                       PVFS_TYPE_METAFILE) {
+               pvfs2_inode->blksize =
+                       new_op->downcall.resp.getattr.attributes.blksize;
+       } else {
+               /* mimic behavior of generic_fillattr() for other types. */
+               pvfs2_inode->blksize = (1 << inode->i_blkbits);
+
+       }
+
+out:
+       gossip_debug(GOSSIP_UTILS_DEBUG,
+                    "Getattr on handle %pU, "
+                    "fsid %d\n  (inode ct = %d) returned %d\n",
+                    &pvfs2_inode->refn.khandle,
+                    pvfs2_inode->refn.fs_id,
+                    (int)atomic_read(&inode->i_count),
+                    ret);
+
+       op_release(new_op);
+       return ret;
+}
+
+/*
+ * issues a pvfs2 setattr request to make sure the new attribute values
+ * take effect if successful.  returns 0 on success; -errno otherwise
+ */
+int pvfs2_inode_setattr(struct inode *inode, struct iattr *iattr)
+{
+       struct pvfs2_inode_s *pvfs2_inode = PVFS2_I(inode);
+       struct pvfs2_kernel_op_s *new_op;
+       int ret;
+
+       new_op = op_alloc(PVFS2_VFS_OP_SETATTR);
+       if (!new_op)
+               return -ENOMEM;
+
+       new_op->upcall.req.setattr.refn = pvfs2_inode->refn;
+       ret = copy_attributes_from_inode(inode,
+                      &new_op->upcall.req.setattr.attributes,
+                      iattr);
+       if (ret < 0) {
+               op_release(new_op);
+               return ret;
+       }
+
+       ret = service_operation(new_op, __func__,
+                               get_interruptible_flag(inode));
+
+       gossip_debug(GOSSIP_UTILS_DEBUG,
+                    "pvfs2_inode_setattr: returning %d\n",
+                    ret);
+
+       /* when request is serviced properly, free req op struct */
+       op_release(new_op);
+
+       /*
+        * successful setattr should clear the atime, mtime and
+        * ctime flags.
+        */
+       if (ret == 0) {
+               ClearAtimeFlag(pvfs2_inode);
+               ClearMtimeFlag(pvfs2_inode);
+               ClearCtimeFlag(pvfs2_inode);
+               ClearModeFlag(pvfs2_inode);
+       }
+
+       return ret;
+}
+
+int pvfs2_flush_inode(struct inode *inode)
+{
+       /*
+        * If it is a dirty inode, this function gets called.
+        * Gather all the information that needs to be setattr'ed
+        * Right now, this will only be used for mode, atime, mtime
+        * and/or ctime.
+        */
+       struct iattr wbattr;
+       int ret;
+       int mtime_flag;
+       int ctime_flag;
+       int atime_flag;
+       int mode_flag;
+       struct pvfs2_inode_s *pvfs2_inode = PVFS2_I(inode);
+
+       memset(&wbattr, 0, sizeof(wbattr));
+
+       /*
+        * check inode flags up front, and clear them if they are set.  This
+        * will prevent multiple processes from all trying to flush the same
+        * inode if they call close() simultaneously
+        */
+       mtime_flag = MtimeFlag(pvfs2_inode);
+       ClearMtimeFlag(pvfs2_inode);
+       ctime_flag = CtimeFlag(pvfs2_inode);
+       ClearCtimeFlag(pvfs2_inode);
+       atime_flag = AtimeFlag(pvfs2_inode);
+       ClearAtimeFlag(pvfs2_inode);
+       mode_flag = ModeFlag(pvfs2_inode);
+       ClearModeFlag(pvfs2_inode);
+
+       /*  -- Lazy atime,mtime and ctime update --
+        * Note: all times are dictated by server in the new scheme
+        * and not by the clients
+        *
+        * Also mode updates are being handled now..
+        */
+
+       if (mtime_flag)
+               wbattr.ia_valid |= ATTR_MTIME;
+       if (ctime_flag)
+               wbattr.ia_valid |= ATTR_CTIME;
+       if (atime_flag)
+               wbattr.ia_valid |= ATTR_ATIME;
+
+       if (mode_flag) {
+               wbattr.ia_mode = inode->i_mode;
+               wbattr.ia_valid |= ATTR_MODE;
+       }
+
+       gossip_debug(GOSSIP_UTILS_DEBUG,
+                    "*********** pvfs2_flush_inode: %pU "
+                    "(ia_valid %d)\n",
+                    get_khandle_from_ino(inode),
+                    wbattr.ia_valid);
+       if (wbattr.ia_valid == 0) {
+               gossip_debug(GOSSIP_UTILS_DEBUG,
+                            "pvfs2_flush_inode skipping setattr()\n");
+               return 0;
+       }
+
+       gossip_debug(GOSSIP_UTILS_DEBUG,
+                    "pvfs2_flush_inode (%pU) writing mode %o\n",
+                    get_khandle_from_ino(inode),
+                    inode->i_mode);
+
+       ret = pvfs2_inode_setattr(inode, &wbattr);
+
+       return ret;
+}
+
+int pvfs2_unmount_sb(struct super_block *sb)
+{
+       int ret = -EINVAL;
+       struct pvfs2_kernel_op_s *new_op = NULL;
+
+       gossip_debug(GOSSIP_UTILS_DEBUG,
+                    "pvfs2_unmount_sb called on sb %p\n",
+                    sb);
+
+       new_op = op_alloc(PVFS2_VFS_OP_FS_UMOUNT);
+       if (!new_op)
+               return -ENOMEM;
+       new_op->upcall.req.fs_umount.id = PVFS2_SB(sb)->id;
+       new_op->upcall.req.fs_umount.fs_id = PVFS2_SB(sb)->fs_id;
+       strncpy(new_op->upcall.req.fs_umount.pvfs2_config_server,
+               PVFS2_SB(sb)->devname,
+               PVFS_MAX_SERVER_ADDR_LEN);
+
+       gossip_debug(GOSSIP_UTILS_DEBUG,
+                    "Attempting PVFS2 Unmount via host %s\n",
+                    new_op->upcall.req.fs_umount.pvfs2_config_server);
+
+       ret = service_operation(new_op, "pvfs2_fs_umount", 0);
+
+       gossip_debug(GOSSIP_UTILS_DEBUG,
+                    "pvfs2_unmount: got return value of %d\n", ret);
+       if (ret)
+               sb = ERR_PTR(ret);
+       else
+               PVFS2_SB(sb)->mount_pending = 1;
+
+       op_release(new_op);
+       return ret;
+}
+
+/*
+ * NOTE: on successful cancellation, be sure to return -EINTR, as
+ * that's the return value the caller expects
+ */
+int pvfs2_cancel_op_in_progress(__u64 tag)
+{
+       int ret = -EINVAL;
+       struct pvfs2_kernel_op_s *new_op = NULL;
+
+       gossip_debug(GOSSIP_UTILS_DEBUG,
+                    "pvfs2_cancel_op_in_progress called on tag %llu\n",
+                    llu(tag));
+
+       new_op = op_alloc(PVFS2_VFS_OP_CANCEL);
+       if (!new_op)
+               return -ENOMEM;
+       new_op->upcall.req.cancel.op_tag = tag;
+
+       gossip_debug(GOSSIP_UTILS_DEBUG,
+                    "Attempting PVFS2 operation cancellation of tag %llu\n",
+                    llu(new_op->upcall.req.cancel.op_tag));
+
+       ret = service_operation(new_op, "pvfs2_cancel", PVFS2_OP_CANCELLATION);
+
+       gossip_debug(GOSSIP_UTILS_DEBUG,
+                    "pvfs2_cancel_op_in_progress: got return value of %d\n",
+                    ret);
+
+       op_release(new_op);
+       return ret;
+}
+
+void pvfs2_op_initialize(struct pvfs2_kernel_op_s *op)
+{
+       if (op) {
+               spin_lock(&op->lock);
+               op->io_completed = 0;
+
+               op->upcall.type = PVFS2_VFS_OP_INVALID;
+               op->downcall.type = PVFS2_VFS_OP_INVALID;
+               op->downcall.status = -1;
+
+               op->op_state = OP_VFS_STATE_UNKNOWN;
+               op->tag = 0;
+               spin_unlock(&op->lock);
+       }
+}
+
+void pvfs2_make_bad_inode(struct inode *inode)
+{
+       if (is_root_handle(inode)) {
+               /*
+                * if this occurs, the pvfs2-client-core was killed but we
+                * can't afford to lose the inode operations and such
+                * associated with the root handle in any case.
+                */
+               gossip_debug(GOSSIP_UTILS_DEBUG,
+                            "*** NOT making bad root inode %pU\n",
+                            get_khandle_from_ino(inode));
+       } else {
+               gossip_debug(GOSSIP_UTILS_DEBUG,
+                            "*** making bad inode %pU\n",
+                            get_khandle_from_ino(inode));
+               make_bad_inode(inode);
+       }
+}
+
+/* this code is based on linux/net/sunrpc/clnt.c:rpc_clnt_sigmask */
+void mask_blocked_signals(sigset_t *orig_sigset)
+{
+       unsigned long sigallow = sigmask(SIGKILL);
+       unsigned long irqflags = 0;
+       struct k_sigaction *action = pvfs2_current_sigaction;
+
+       sigallow |= ((action[SIGINT - 1].sa.sa_handler == SIG_DFL) ?
+                    sigmask(SIGINT) :
+                    0);
+       sigallow |= ((action[SIGQUIT - 1].sa.sa_handler == SIG_DFL) ?
+                    sigmask(SIGQUIT) :
+                    0);
+
+       spin_lock_irqsave(&pvfs2_current_signal_lock, irqflags);
+       *orig_sigset = current->blocked;
+       siginitsetinv(&current->blocked, sigallow & ~orig_sigset->sig[0]);
+       recalc_sigpending();
+       spin_unlock_irqrestore(&pvfs2_current_signal_lock, irqflags);
+}
+
+/* this code is based on linux/net/sunrpc/clnt.c:rpc_clnt_sigunmask */
+void unmask_blocked_signals(sigset_t *orig_sigset)
+{
+       unsigned long irqflags = 0;
+
+       spin_lock_irqsave(&pvfs2_current_signal_lock, irqflags);
+       current->blocked = *orig_sigset;
+       recalc_sigpending();
+       spin_unlock_irqrestore(&pvfs2_current_signal_lock, irqflags);
+}
+
+__u64 pvfs2_convert_time_field(void *time_ptr)
+{
+       __u64 pvfs2_time;
+       struct timespec *tspec = (struct timespec *)time_ptr;
+
+       pvfs2_time = (__u64) ((time_t) tspec->tv_sec);
+       return pvfs2_time;
+}
+
+/* macro defined in include/pvfs2-types.h */
+DECLARE_ERRNO_MAPPING_AND_FN();
+
+int pvfs2_normalize_to_errno(__s32 error_code)
+{
+       if (error_code > 0) {
+               gossip_err("pvfs2: error status receieved.\n");
+               gossip_err("pvfs2: assuming error code is inverted.\n");
+               error_code = -error_code;
+       }
+
+       /* convert any error codes that are in pvfs2 format */
+       if (IS_PVFS_NON_ERRNO_ERROR(-error_code)) {
+               if (PVFS_NON_ERRNO_ERROR_CODE(-error_code) == PVFS_ECANCEL) {
+                       /*
+                        * cancellation error codes generally correspond to
+                        * a timeout from the client's perspective
+                        */
+                       error_code = -ETIMEDOUT;
+               } else {
+                       /* assume a default error code */
+                       gossip_err("pvfs2: warning: got error code without errno equivalent: %d.\n",
+                                  error_code);
+                       error_code = -EINVAL;
+               }
+       } else if (IS_PVFS_ERROR(-error_code)) {
+               error_code = -PVFS_ERROR_TO_ERRNO(-error_code);
+       }
+       return error_code;
+}
+
+#define NUM_MODES 11
+__s32 PVFS_util_translate_mode(int mode)
+{
+       int ret = 0;
+       int i = 0;
+       static int modes[NUM_MODES] = {
+               S_IXOTH, S_IWOTH, S_IROTH,
+               S_IXGRP, S_IWGRP, S_IRGRP,
+               S_IXUSR, S_IWUSR, S_IRUSR,
+               S_ISGID, S_ISUID
+       };
+       static int pvfs2_modes[NUM_MODES] = {
+               PVFS_O_EXECUTE, PVFS_O_WRITE, PVFS_O_READ,
+               PVFS_G_EXECUTE, PVFS_G_WRITE, PVFS_G_READ,
+               PVFS_U_EXECUTE, PVFS_U_WRITE, PVFS_U_READ,
+               PVFS_G_SGID, PVFS_U_SUID
+       };
+
+       for (i = 0; i < NUM_MODES; i++)
+               if (mode & modes[i])
+                       ret |= pvfs2_modes[i];
+
+       return ret;
+}
+#undef NUM_MODES
+
+/*
+ * After obtaining a string representation of the client's debug
+ * keywords and their associated masks, this function is called to build an
+ * array of these values.
+ */
+int orangefs_prepare_cdm_array(char *debug_array_string)
+{
+       int i;
+       int rc = -EINVAL;
+       char *cds_head = NULL;
+       char *cds_delimiter = NULL;
+       int keyword_len = 0;
+
+       gossip_debug(GOSSIP_UTILS_DEBUG, "%s: start\n", __func__);
+
+       /*
+        * figure out how many elements the cdm_array needs.
+        */
+       for (i = 0; i < strlen(debug_array_string); i++)
+               if (debug_array_string[i] == '\n')
+                       cdm_element_count++;
+
+       if (!cdm_element_count) {
+               pr_info("No elements in client debug array string!\n");
+               goto out;
+       }
+
+       cdm_array =
+               kzalloc(cdm_element_count * sizeof(struct client_debug_mask),
+                       GFP_KERNEL);
+       if (!cdm_array) {
+               pr_info("malloc failed for cdm_array!\n");
+               rc = -ENOMEM;
+               goto out;
+       }
+
+       cds_head = debug_array_string;
+
+       for (i = 0; i < cdm_element_count; i++) {
+               cds_delimiter = strchr(cds_head, '\n');
+               *cds_delimiter = '\0';
+
+               keyword_len = strcspn(cds_head, " ");
+
+               cdm_array[i].keyword = kzalloc(keyword_len + 1, GFP_KERNEL);
+               if (!cdm_array[i].keyword) {
+                       rc = -ENOMEM;
+                       goto out;
+               }
+
+               sscanf(cds_head,
+                      "%s %llx %llx",
+                      cdm_array[i].keyword,
+                      (unsigned long long *)&(cdm_array[i].mask1),
+                      (unsigned long long *)&(cdm_array[i].mask2));
+
+               if (!strcmp(cdm_array[i].keyword, PVFS2_VERBOSE))
+                       client_verbose_index = i;
+
+               if (!strcmp(cdm_array[i].keyword, PVFS2_ALL))
+                       client_all_index = i;
+
+               cds_head = cds_delimiter + 1;
+       }
+
+       rc = cdm_element_count;
+
+       gossip_debug(GOSSIP_UTILS_DEBUG, "%s: rc:%d:\n", __func__, rc);
+
+out:
+
+       return rc;
+
+}
+
+/*
+ * /sys/kernel/debug/orangefs/debug-help can be catted to
+ * see all the available kernel and client debug keywords.
+ *
+ * When the kernel boots, we have no idea what keywords the
+ * client supports, nor their associated masks.
+ *
+ * We pass through this function once at boot and stamp a
+ * boilerplate "we don't know" message for the client in the
+ * debug-help file. We pass through here again when the client
+ * starts and then we can fill out the debug-help file fully.
+ *
+ * The client might be restarted any number of times between
+ * reboots, we only build the debug-help file the first time.
+ */
+int orangefs_prepare_debugfs_help_string(int at_boot)
+{
+       int rc = -EINVAL;
+       int i;
+       int byte_count = 0;
+       char *client_title = "Client Debug Keywords:\n";
+       char *kernel_title = "Kernel Debug Keywords:\n";
+
+       gossip_debug(GOSSIP_UTILS_DEBUG, "%s: start\n", __func__);
+
+       if (at_boot) {
+               byte_count += strlen(HELP_STRING_UNINITIALIZED);
+               client_title = HELP_STRING_UNINITIALIZED;
+       } else {
+               /*
+                * fill the client keyword/mask array and remember
+                * how many elements there were.
+                */
+               cdm_element_count =
+                       orangefs_prepare_cdm_array(client_debug_array_string);
+               if (cdm_element_count <= 0)
+                       goto out;
+
+               /* Count the bytes destined for debug_help_string. */
+               byte_count += strlen(client_title);
+
+               for (i = 0; i < cdm_element_count; i++) {
+                       byte_count += strlen(cdm_array[i].keyword + 2);
+                       if (byte_count >= DEBUG_HELP_STRING_SIZE) {
+                               pr_info("%s: overflow 1!\n", __func__);
+                               goto out;
+                       }
+               }
+
+               gossip_debug(GOSSIP_UTILS_DEBUG,
+                            "%s: cdm_element_count:%d:\n",
+                            __func__,
+                            cdm_element_count);
+       }
+
+       byte_count += strlen(kernel_title);
+       for (i = 0; i < num_kmod_keyword_mask_map; i++) {
+               byte_count +=
+                       strlen(s_kmod_keyword_mask_map[i].keyword + 2);
+               if (byte_count >= DEBUG_HELP_STRING_SIZE) {
+                       pr_info("%s: overflow 2!\n", __func__);
+                       goto out;
+               }
+       }
+
+       /* build debug_help_string. */
+       debug_help_string = kzalloc(DEBUG_HELP_STRING_SIZE, GFP_KERNEL);
+       if (!debug_help_string) {
+               rc = -ENOMEM;
+               goto out;
+       }
+
+       strcat(debug_help_string, client_title);
+
+       if (!at_boot) {
+               for (i = 0; i < cdm_element_count; i++) {
+                       strcat(debug_help_string, "\t");
+                       strcat(debug_help_string, cdm_array[i].keyword);
+                       strcat(debug_help_string, "\n");
+               }
+       }
+
+       strcat(debug_help_string, "\n");
+       strcat(debug_help_string, kernel_title);
+
+       for (i = 0; i < num_kmod_keyword_mask_map; i++) {
+               strcat(debug_help_string, "\t");
+               strcat(debug_help_string, s_kmod_keyword_mask_map[i].keyword);
+               strcat(debug_help_string, "\n");
+       }
+
+       rc = 0;
+
+out:
+
+       return rc;
+
+}
+
+/*
+ * kernel = type 0
+ * client = type 1
+ */
+void debug_mask_to_string(void *mask, int type)
+{
+       int i;
+       int len = 0;
+       char *debug_string;
+       int element_count = 0;
+
+       gossip_debug(GOSSIP_UTILS_DEBUG, "%s: start\n", __func__);
+
+       if (type) {
+               debug_string = client_debug_string;
+               element_count = cdm_element_count;
+       } else {
+               debug_string = kernel_debug_string;
+               element_count = num_kmod_keyword_mask_map;
+       }
+
+       memset(debug_string, 0, PVFS2_MAX_DEBUG_STRING_LEN);
+
+       /*
+        * Some keywords, like "all" or "verbose", are amalgams of
+        * numerous other keywords. Make a special check for those
+        * before grinding through the whole mask only to find out
+        * later...
+        */
+       if (check_amalgam_keyword(mask, type))
+               goto out;
+
+       /* Build the debug string. */
+       for (i = 0; i < element_count; i++)
+               if (type)
+                       do_c_string(mask, i);
+               else
+                       do_k_string(mask, i);
+
+       len = strlen(debug_string);
+
+       if ((len) && (type))
+               client_debug_string[len - 1] = '\0';
+       else if (len)
+               kernel_debug_string[len - 1] = '\0';
+       else if (type)
+               strcpy(client_debug_string, "none");
+       else
+               strcpy(kernel_debug_string, "none");
+
+out:
+gossip_debug(GOSSIP_UTILS_DEBUG, "%s: string:%s:\n", __func__, debug_string);
+
+       return;
+
+}
+
+void do_k_string(void *k_mask, int index)
+{
+       __u64 *mask = (__u64 *) k_mask;
+
+       if (keyword_is_amalgam((char *) s_kmod_keyword_mask_map[index].keyword))
+                       goto out;
+
+       if (*mask & s_kmod_keyword_mask_map[index].mask_val) {
+               if ((strlen(kernel_debug_string) +
+                    strlen(s_kmod_keyword_mask_map[index].keyword))
+                       < PVFS2_MAX_DEBUG_STRING_LEN - 1) {
+                               strcat(kernel_debug_string,
+                                      s_kmod_keyword_mask_map[index].keyword);
+                               strcat(kernel_debug_string, ",");
+                       } else {
+                               gossip_err("%s: overflow!\n", __func__);
+                               strcpy(kernel_debug_string, PVFS2_ALL);
+                               goto out;
+                       }
+       }
+
+out:
+
+       return;
+}
+
+void do_c_string(void *c_mask, int index)
+{
+       struct client_debug_mask *mask = (struct client_debug_mask *) c_mask;
+
+       if (keyword_is_amalgam(cdm_array[index].keyword))
+               goto out;
+
+       if ((mask->mask1 & cdm_array[index].mask1) ||
+           (mask->mask2 & cdm_array[index].mask2)) {
+               if ((strlen(client_debug_string) +
+                    strlen(cdm_array[index].keyword) + 1)
+                       < PVFS2_MAX_DEBUG_STRING_LEN - 2) {
+                               strcat(client_debug_string,
+                                      cdm_array[index].keyword);
+                               strcat(client_debug_string, ",");
+                       } else {
+                               gossip_err("%s: overflow!\n", __func__);
+                               strcpy(client_debug_string, PVFS2_ALL);
+                               goto out;
+                       }
+       }
+out:
+       return;
+}
+
+int keyword_is_amalgam(char *keyword)
+{
+       int rc = 0;
+
+       if ((!strcmp(keyword, PVFS2_ALL)) || (!strcmp(keyword, PVFS2_VERBOSE)))
+               rc = 1;
+
+       return rc;
+}
+
+/*
+ * kernel = type 0
+ * client = type 1
+ *
+ * return 1 if we found an amalgam.
+ */
+int check_amalgam_keyword(void *mask, int type)
+{
+       __u64 *k_mask;
+       struct client_debug_mask *c_mask;
+       int k_all_index = num_kmod_keyword_mask_map - 1;
+       int rc = 0;
+
+       if (type) {
+               c_mask = (struct client_debug_mask *) mask;
+
+               if ((c_mask->mask1 == cdm_array[client_all_index].mask1) &&
+                   (c_mask->mask2 == cdm_array[client_all_index].mask2)) {
+                       strcpy(client_debug_string, PVFS2_ALL);
+                       rc = 1;
+                       goto out;
+               }
+
+               if ((c_mask->mask1 == cdm_array[client_verbose_index].mask1) &&
+                   (c_mask->mask2 == cdm_array[client_verbose_index].mask2)) {
+                       strcpy(client_debug_string, PVFS2_VERBOSE);
+                       rc = 1;
+                       goto out;
+               }
+
+       } else {
+               k_mask = (__u64 *) mask;
+
+               if (*k_mask >= s_kmod_keyword_mask_map[k_all_index].mask_val) {
+                       strcpy(kernel_debug_string, PVFS2_ALL);
+                       rc = 1;
+                       goto out;
+               }
+       }
+
+out:
+
+       return rc;
+}
+
+/*
+ * kernel = type 0
+ * client = type 1
+ */
+void debug_string_to_mask(char *debug_string, void *mask, int type)
+{
+       char *unchecked_keyword;
+       int i;
+       char *strsep_fodder = kstrdup(debug_string, GFP_KERNEL);
+       int element_count = 0;
+       struct client_debug_mask *c_mask;
+       __u64 *k_mask;
+
+       gossip_debug(GOSSIP_UTILS_DEBUG, "%s: start\n", __func__);
+
+       if (type) {
+               c_mask = (struct client_debug_mask *)mask;
+               element_count = cdm_element_count;
+       } else {
+               k_mask = (__u64 *)mask;
+               *k_mask = 0;
+               element_count = num_kmod_keyword_mask_map;
+       }
+
+       while ((unchecked_keyword = strsep(&strsep_fodder, ",")))
+               if (strlen(unchecked_keyword)) {
+                       for (i = 0; i < element_count; i++)
+                               if (type)
+                                       do_c_mask(i,
+                                                 unchecked_keyword,
+                                                 &c_mask);
+                               else
+                                       do_k_mask(i,
+                                                 unchecked_keyword,
+                                                 &k_mask);
+               }
+
+       kfree(strsep_fodder);
+}
+
+void do_c_mask(int i,
+              char *unchecked_keyword,
+              struct client_debug_mask **sane_mask)
+{
+
+       if (!strcmp(cdm_array[i].keyword, unchecked_keyword)) {
+               (**sane_mask).mask1 = (**sane_mask).mask1 | cdm_array[i].mask1;
+               (**sane_mask).mask2 = (**sane_mask).mask2 | cdm_array[i].mask2;
+       }
+}
+
+void do_k_mask(int i, char *unchecked_keyword, __u64 **sane_mask)
+{
+
+       if (!strcmp(s_kmod_keyword_mask_map[i].keyword, unchecked_keyword))
+               **sane_mask = (**sane_mask) |
+                               s_kmod_keyword_mask_map[i].mask_val;
+}