Add the rt linux 4.1.3-rt3 as base
[kvmfornfv.git] / kernel / drivers / md / dm-log-userspace-transfer.c
diff --git a/kernel/drivers/md/dm-log-userspace-transfer.c b/kernel/drivers/md/dm-log-userspace-transfer.c
new file mode 100644 (file)
index 0000000..fdf8ec3
--- /dev/null
@@ -0,0 +1,287 @@
+/*
+ * Copyright (C) 2006-2009 Red Hat, Inc.
+ *
+ * This file is released under the LGPL.
+ */
+
+#include <linux/kernel.h>
+#include <linux/module.h>
+#include <linux/slab.h>
+#include <net/sock.h>
+#include <linux/workqueue.h>
+#include <linux/connector.h>
+#include <linux/device-mapper.h>
+#include <linux/dm-log-userspace.h>
+
+#include "dm-log-userspace-transfer.h"
+
+static uint32_t dm_ulog_seq;
+
+/*
+ * Netlink/Connector is an unreliable protocol.  How long should
+ * we wait for a response before assuming it was lost and retrying?
+ * (If we do receive a response after this time, it will be discarded
+ * and the response to the resent request will be waited for.
+ */
+#define DM_ULOG_RETRY_TIMEOUT (15 * HZ)
+
+/*
+ * Pre-allocated space for speed
+ */
+#define DM_ULOG_PREALLOCED_SIZE 512
+static struct cn_msg *prealloced_cn_msg;
+static struct dm_ulog_request *prealloced_ulog_tfr;
+
+static struct cb_id ulog_cn_id = {
+       .idx = CN_IDX_DM,
+       .val = CN_VAL_DM_USERSPACE_LOG
+};
+
+static DEFINE_MUTEX(dm_ulog_lock);
+
+struct receiving_pkg {
+       struct list_head list;
+       struct completion complete;
+
+       uint32_t seq;
+
+       int error;
+       size_t *data_size;
+       char *data;
+};
+
+static DEFINE_SPINLOCK(receiving_list_lock);
+static struct list_head receiving_list;
+
+static int dm_ulog_sendto_server(struct dm_ulog_request *tfr)
+{
+       int r;
+       struct cn_msg *msg = prealloced_cn_msg;
+
+       memset(msg, 0, sizeof(struct cn_msg));
+
+       msg->id.idx = ulog_cn_id.idx;
+       msg->id.val = ulog_cn_id.val;
+       msg->ack = 0;
+       msg->seq = tfr->seq;
+       msg->len = sizeof(struct dm_ulog_request) + tfr->data_size;
+
+       r = cn_netlink_send(msg, 0, 0, gfp_any());
+
+       return r;
+}
+
+/*
+ * Parameters for this function can be either msg or tfr, but not
+ * both.  This function fills in the reply for a waiting request.
+ * If just msg is given, then the reply is simply an ACK from userspace
+ * that the request was received.
+ *
+ * Returns: 0 on success, -ENOENT on failure
+ */
+static int fill_pkg(struct cn_msg *msg, struct dm_ulog_request *tfr)
+{
+       uint32_t rtn_seq = (msg) ? msg->seq : (tfr) ? tfr->seq : 0;
+       struct receiving_pkg *pkg;
+
+       /*
+        * The 'receiving_pkg' entries in this list are statically
+        * allocated on the stack in 'dm_consult_userspace'.
+        * Each process that is waiting for a reply from the user
+        * space server will have an entry in this list.
+        *
+        * We are safe to do it this way because the stack space
+        * is unique to each process, but still addressable by
+        * other processes.
+        */
+       list_for_each_entry(pkg, &receiving_list, list) {
+               if (rtn_seq != pkg->seq)
+                       continue;
+
+               if (msg) {
+                       pkg->error = -msg->ack;
+                       /*
+                        * If we are trying again, we will need to know our
+                        * storage capacity.  Otherwise, along with the
+                        * error code, we make explicit that we have no data.
+                        */
+                       if (pkg->error != -EAGAIN)
+                               *(pkg->data_size) = 0;
+               } else if (tfr->data_size > *(pkg->data_size)) {
+                       DMERR("Insufficient space to receive package [%u] "
+                             "(%u vs %zu)", tfr->request_type,
+                             tfr->data_size, *(pkg->data_size));
+
+                       *(pkg->data_size) = 0;
+                       pkg->error = -ENOSPC;
+               } else {
+                       pkg->error = tfr->error;
+                       memcpy(pkg->data, tfr->data, tfr->data_size);
+                       *(pkg->data_size) = tfr->data_size;
+               }
+               complete(&pkg->complete);
+               return 0;
+       }
+
+       return -ENOENT;
+}
+
+/*
+ * This is the connector callback that delivers data
+ * that was sent from userspace.
+ */
+static void cn_ulog_callback(struct cn_msg *msg, struct netlink_skb_parms *nsp)
+{
+       struct dm_ulog_request *tfr = (struct dm_ulog_request *)(msg + 1);
+
+       if (!capable(CAP_SYS_ADMIN))
+               return;
+
+       spin_lock(&receiving_list_lock);
+       if (msg->len == 0)
+               fill_pkg(msg, NULL);
+       else if (msg->len < sizeof(*tfr))
+               DMERR("Incomplete message received (expected %u, got %u): [%u]",
+                     (unsigned)sizeof(*tfr), msg->len, msg->seq);
+       else
+               fill_pkg(NULL, tfr);
+       spin_unlock(&receiving_list_lock);
+}
+
+/**
+ * dm_consult_userspace
+ * @uuid: log's universal unique identifier (must be DM_UUID_LEN in size)
+ * @luid: log's local unique identifier
+ * @request_type:  found in include/linux/dm-log-userspace.h
+ * @data: data to tx to the server
+ * @data_size: size of data in bytes
+ * @rdata: place to put return data from server
+ * @rdata_size: value-result (amount of space given/amount of space used)
+ *
+ * rdata_size is undefined on failure.
+ *
+ * Memory used to communicate with userspace is zero'ed
+ * before populating to ensure that no unwanted bits leak
+ * from kernel space to user-space.  All userspace log communications
+ * between kernel and user space go through this function.
+ *
+ * Returns: 0 on success, -EXXX on failure
+ **/
+int dm_consult_userspace(const char *uuid, uint64_t luid, int request_type,
+                        char *data, size_t data_size,
+                        char *rdata, size_t *rdata_size)
+{
+       int r = 0;
+       unsigned long tmo;
+       size_t dummy = 0;
+       int overhead_size = sizeof(struct dm_ulog_request) + sizeof(struct cn_msg);
+       struct dm_ulog_request *tfr = prealloced_ulog_tfr;
+       struct receiving_pkg pkg;
+
+       /*
+        * Given the space needed to hold the 'struct cn_msg' and
+        * 'struct dm_ulog_request' - do we have enough payload
+        * space remaining?
+        */
+       if (data_size > (DM_ULOG_PREALLOCED_SIZE - overhead_size)) {
+               DMINFO("Size of tfr exceeds preallocated size");
+               return -EINVAL;
+       }
+
+       if (!rdata_size)
+               rdata_size = &dummy;
+resend:
+       /*
+        * We serialize the sending of requests so we can
+        * use the preallocated space.
+        */
+       mutex_lock(&dm_ulog_lock);
+
+       memset(tfr, 0, DM_ULOG_PREALLOCED_SIZE - sizeof(struct cn_msg));
+       memcpy(tfr->uuid, uuid, DM_UUID_LEN);
+       tfr->version = DM_ULOG_REQUEST_VERSION;
+       tfr->luid = luid;
+       tfr->seq = dm_ulog_seq++;
+
+       /*
+        * Must be valid request type (all other bits set to
+        * zero).  This reserves other bits for possible future
+        * use.
+        */
+       tfr->request_type = request_type & DM_ULOG_REQUEST_MASK;
+
+       tfr->data_size = data_size;
+       if (data && data_size)
+               memcpy(tfr->data, data, data_size);
+
+       memset(&pkg, 0, sizeof(pkg));
+       init_completion(&pkg.complete);
+       pkg.seq = tfr->seq;
+       pkg.data_size = rdata_size;
+       pkg.data = rdata;
+       spin_lock(&receiving_list_lock);
+       list_add(&(pkg.list), &receiving_list);
+       spin_unlock(&receiving_list_lock);
+
+       r = dm_ulog_sendto_server(tfr);
+
+       mutex_unlock(&dm_ulog_lock);
+
+       if (r) {
+               DMERR("Unable to send log request [%u] to userspace: %d",
+                     request_type, r);
+               spin_lock(&receiving_list_lock);
+               list_del_init(&(pkg.list));
+               spin_unlock(&receiving_list_lock);
+
+               goto out;
+       }
+
+       tmo = wait_for_completion_timeout(&(pkg.complete), DM_ULOG_RETRY_TIMEOUT);
+       spin_lock(&receiving_list_lock);
+       list_del_init(&(pkg.list));
+       spin_unlock(&receiving_list_lock);
+       if (!tmo) {
+               DMWARN("[%s] Request timed out: [%u/%u] - retrying",
+                      (strlen(uuid) > 8) ?
+                      (uuid + (strlen(uuid) - 8)) : (uuid),
+                      request_type, pkg.seq);
+               goto resend;
+       }
+
+       r = pkg.error;
+       if (r == -EAGAIN)
+               goto resend;
+
+out:
+       return r;
+}
+
+int dm_ulog_tfr_init(void)
+{
+       int r;
+       void *prealloced;
+
+       INIT_LIST_HEAD(&receiving_list);
+
+       prealloced = kmalloc(DM_ULOG_PREALLOCED_SIZE, GFP_KERNEL);
+       if (!prealloced)
+               return -ENOMEM;
+
+       prealloced_cn_msg = prealloced;
+       prealloced_ulog_tfr = prealloced + sizeof(struct cn_msg);
+
+       r = cn_add_callback(&ulog_cn_id, "dmlogusr", cn_ulog_callback);
+       if (r) {
+               kfree(prealloced_cn_msg);
+               return r;
+       }
+
+       return 0;
+}
+
+void dm_ulog_tfr_exit(void)
+{
+       cn_del_callback(&ulog_cn_id);
+       kfree(prealloced_cn_msg);
+}