--- /dev/null
+/*
+ * osd_initiator - Main body of the osd initiator library.
+ *
+ * Note: The file does not contain the advanced security functionality which
+ * is only needed by the security_manager's initiators.
+ *
+ * Copyright (C) 2008 Panasas Inc.  All rights reserved.
+ *
+ * Authors:
+ *   Boaz Harrosh <bharrosh@panasas.com>
+ *   Benny Halevy <bhalevy@panasas.com>
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License version 2
+ *
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted provided that the following conditions
+ * are met:
+ *
+ *  1. Redistributions of source code must retain the above copyright
+ *     notice, this list of conditions and the following disclaimer.
+ *  2. Redistributions in binary form must reproduce the above copyright
+ *     notice, this list of conditions and the following disclaimer in the
+ *     documentation and/or other materials provided with the distribution.
+ *  3. Neither the name of the Panasas company nor the names of its
+ *     contributors may be used to endorse or promote products derived
+ *     from this software without specific prior written permission.
+ *
+ * THIS SOFTWARE IS PROVIDED ``AS IS'' AND ANY EXPRESS OR IMPLIED
+ * WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF
+ * MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
+ * DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
+ * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
+ * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
+ * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR
+ * BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
+ * LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
+ * NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
+ * SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+ */
+
+#include <scsi/osd_initiator.h>
+#include <scsi/osd_sec.h>
+#include <scsi/scsi_device.h>
+
+#include "osd_debug.h"
+
+enum { OSD_REQ_RETRIES = 1 };
+
+MODULE_AUTHOR("Boaz Harrosh <bharrosh@panasas.com>");
+MODULE_DESCRIPTION("open-osd initiator library libosd.ko");
+MODULE_LICENSE("GPL");
+
+static inline void build_test(void)
+{
+       /* structures were not packed */
+       BUILD_BUG_ON(sizeof(struct osd_capability) != OSD_CAP_LEN);
+       BUILD_BUG_ON(sizeof(struct osdv1_cdb) != OSDv1_TOTAL_CDB_LEN);
+}
+
+static unsigned _osd_req_cdb_len(struct osd_request *or)
+{
+       return OSDv1_TOTAL_CDB_LEN;
+}
+
+void osd_dev_init(struct osd_dev *osdd, struct scsi_device *scsi_device)
+{
+       memset(osdd, 0, sizeof(*osdd));
+       osdd->scsi_device = scsi_device;
+       osdd->def_timeout = BLK_DEFAULT_SG_TIMEOUT;
+       /* TODO: Allocate pools for osd_request attributes ... */
+}
+EXPORT_SYMBOL(osd_dev_init);
+
+void osd_dev_fini(struct osd_dev *osdd)
+{
+       /* TODO: De-allocate pools */
+
+       osdd->scsi_device = NULL;
+}
+EXPORT_SYMBOL(osd_dev_fini);
+
+static struct osd_request *_osd_request_alloc(gfp_t gfp)
+{
+       struct osd_request *or;
+
+       /* TODO: Use mempool with one saved request */
+       or = kzalloc(sizeof(*or), gfp);
+       return or;
+}
+
+static void _osd_request_free(struct osd_request *or)
+{
+       kfree(or);
+}
+
+struct osd_request *osd_start_request(struct osd_dev *dev, gfp_t gfp)
+{
+       struct osd_request *or;
+
+       or = _osd_request_alloc(gfp);
+       if (!or)
+               return NULL;
+
+       or->osd_dev = dev;
+       or->alloc_flags = gfp;
+       or->timeout = dev->def_timeout;
+       or->retries = OSD_REQ_RETRIES;
+
+       return or;
+}
+EXPORT_SYMBOL(osd_start_request);
+
+/*
+ * If osd_finalize_request() was called but the request was not executed through
+ * the block layer, then we must release BIOs.
+ */
+static void _abort_unexecuted_bios(struct request *rq)
+{
+       struct bio *bio;
+
+       while ((bio = rq->bio) != NULL) {
+               rq->bio = bio->bi_next;
+               bio_endio(bio, 0);
+       }
+}
+
+void osd_end_request(struct osd_request *or)
+{
+       struct request *rq = or->request;
+
+       if (rq) {
+               if (rq->next_rq) {
+                       _abort_unexecuted_bios(rq->next_rq);
+                       blk_put_request(rq->next_rq);
+               }
+
+               _abort_unexecuted_bios(rq);
+               blk_put_request(rq);
+       }
+       _osd_request_free(or);
+}
+EXPORT_SYMBOL(osd_end_request);
+
+int osd_execute_request(struct osd_request *or)
+{
+       return blk_execute_rq(or->request->q, NULL, or->request, 0);
+}
+EXPORT_SYMBOL(osd_execute_request);
+
+static void osd_request_async_done(struct request *req, int error)
+{
+       struct osd_request *or = req->end_io_data;
+
+       or->async_error = error;
+
+       if (error)
+               OSD_DEBUG("osd_request_async_done error recieved %d\n", error);
+
+       if (or->async_done)
+               or->async_done(or, or->async_private);
+       else
+               osd_end_request(or);
+}
+
+int osd_execute_request_async(struct osd_request *or,
+       osd_req_done_fn *done, void *private)
+{
+       or->request->end_io_data = or;
+       or->async_private = private;
+       or->async_done = done;
+
+       blk_execute_rq_nowait(or->request->q, NULL, or->request, 0,
+                             osd_request_async_done);
+       return 0;
+}
+EXPORT_SYMBOL(osd_execute_request_async);
+
+/*
+ * Common to all OSD commands
+ */
+
+static void _osdv1_req_encode_common(struct osd_request *or,
+       __be16 act, const struct osd_obj_id *obj, u64 offset, u64 len)
+{
+       struct osdv1_cdb *ocdb = &or->cdb.v1;
+
+       /*
+        * For speed, the commands
+        *      OSD_ACT_PERFORM_SCSI_COMMAND    , V1 0x8F7E, V2 0x8F7C
+        *      OSD_ACT_SCSI_TASK_MANAGEMENT    , V1 0x8F7F, V2 0x8F7D
+        * are not supported here. Should pass zero and set after the call
+        */
+       act &= cpu_to_be16(~0x0080); /* V1 action code */
+
+       OSD_DEBUG("OSDv1 execute opcode 0x%x\n", be16_to_cpu(act));
+
+       ocdb->h.varlen_cdb.opcode = VARIABLE_LENGTH_CMD;
+       ocdb->h.varlen_cdb.additional_cdb_length = OSD_ADDITIONAL_CDB_LENGTH;
+       ocdb->h.varlen_cdb.service_action = act;
+
+       ocdb->h.partition = cpu_to_be64(obj->partition);
+       ocdb->h.object = cpu_to_be64(obj->id);
+       ocdb->h.v1.length = cpu_to_be64(len);
+       ocdb->h.v1.start_address = cpu_to_be64(offset);
+}
+
+static void _osd_req_encode_common(struct osd_request *or,
+       __be16 act, const struct osd_obj_id *obj, u64 offset, u64 len)
+{
+       _osdv1_req_encode_common(or, act, obj, offset, len);
+}
+
+/*
+ * Device commands
+ */
+void osd_req_format(struct osd_request *or, u64 tot_capacity)
+{
+       _osd_req_encode_common(or, OSD_ACT_FORMAT_OSD, &osd_root_object, 0,
+                               tot_capacity);
+}
+EXPORT_SYMBOL(osd_req_format);
+
+/*
+ * Partition commands
+ */
+static void _osd_req_encode_partition(struct osd_request *or,
+       __be16 act, osd_id partition)
+{
+       struct osd_obj_id par = {
+               .partition = partition,
+               .id = 0,
+       };
+
+       _osd_req_encode_common(or, act, &par, 0, 0);
+}
+
+void osd_req_create_partition(struct osd_request *or, osd_id partition)
+{
+       _osd_req_encode_partition(or, OSD_ACT_CREATE_PARTITION, partition);
+}
+EXPORT_SYMBOL(osd_req_create_partition);
+
+void osd_req_remove_partition(struct osd_request *or, osd_id partition)
+{
+       _osd_req_encode_partition(or, OSD_ACT_REMOVE_PARTITION, partition);
+}
+EXPORT_SYMBOL(osd_req_remove_partition);
+
+/*
+ * Object commands
+ */
+void osd_req_create_object(struct osd_request *or, struct osd_obj_id *obj)
+{
+       _osd_req_encode_common(or, OSD_ACT_CREATE, obj, 0, 0);
+}
+EXPORT_SYMBOL(osd_req_create_object);
+
+void osd_req_remove_object(struct osd_request *or, struct osd_obj_id *obj)
+{
+       _osd_req_encode_common(or, OSD_ACT_REMOVE, obj, 0, 0);
+}
+EXPORT_SYMBOL(osd_req_remove_object);
+
+void osd_req_write(struct osd_request *or,
+       const struct osd_obj_id *obj, struct bio *bio, u64 offset)
+{
+       _osd_req_encode_common(or, OSD_ACT_WRITE, obj, offset, bio->bi_size);
+       WARN_ON(or->out.bio || or->out.total_bytes);
+       bio->bi_rw |= (1 << BIO_RW);
+       or->out.bio = bio;
+       or->out.total_bytes = bio->bi_size;
+}
+EXPORT_SYMBOL(osd_req_write);
+
+void osd_req_read(struct osd_request *or,
+       const struct osd_obj_id *obj, struct bio *bio, u64 offset)
+{
+       _osd_req_encode_common(or, OSD_ACT_READ, obj, offset, bio->bi_size);
+       WARN_ON(or->in.bio || or->in.total_bytes);
+       bio->bi_rw &= ~(1 << BIO_RW);
+       or->in.bio = bio;
+       or->in.total_bytes = bio->bi_size;
+}
+EXPORT_SYMBOL(osd_req_read);
+
+/*
+ * osd_finalize_request and helpers
+ */
+
+static int _init_blk_request(struct osd_request *or,
+       bool has_in, bool has_out)
+{
+       gfp_t flags = or->alloc_flags;
+       struct scsi_device *scsi_device = or->osd_dev->scsi_device;
+       struct request_queue *q = scsi_device->request_queue;
+       struct request *req;
+       int ret = -ENOMEM;
+
+       req = blk_get_request(q, has_out, flags);
+       if (!req)
+               goto out;
+
+       or->request = req;
+       req->cmd_type = REQ_TYPE_BLOCK_PC;
+       req->timeout = or->timeout;
+       req->retries = or->retries;
+       req->sense = or->sense;
+       req->sense_len = 0;
+
+       if (has_out) {
+               or->out.req = req;
+               if (has_in) {
+                       /* allocate bidi request */
+                       req = blk_get_request(q, READ, flags);
+                       if (!req) {
+                               OSD_DEBUG("blk_get_request for bidi failed\n");
+                               goto out;
+                       }
+                       req->cmd_type = REQ_TYPE_BLOCK_PC;
+                       or->in.req = or->request->next_rq = req;
+               }
+       } else if (has_in)
+               or->in.req = req;
+
+       ret = 0;
+out:
+       OSD_DEBUG("or=%p has_in=%d has_out=%d => %d, %p\n",
+                       or, has_in, has_out, ret, or->request);
+       return ret;
+}
+
+int osd_finalize_request(struct osd_request *or,
+       u8 options, const void *cap, const u8 *cap_key)
+{
+       struct osd_cdb_head *cdbh = osd_cdb_head(&or->cdb);
+       bool has_in, has_out;
+       int ret;
+
+       if (options & OSD_REQ_FUA)
+               cdbh->options |= OSD_CDB_FUA;
+
+       if (options & OSD_REQ_DPO)
+               cdbh->options |= OSD_CDB_DPO;
+
+       if (options & OSD_REQ_BYPASS_TIMESTAMPS)
+               cdbh->timestamp_control = OSD_CDB_BYPASS_TIMESTAMPS;
+
+       osd_set_caps(&or->cdb, cap);
+
+       has_in = or->in.bio || or->get_attr.total_bytes;
+       has_out = or->out.bio || or->set_attr.total_bytes ||
+               or->enc_get_attr.total_bytes;
+
+       ret = _init_blk_request(or, has_in, has_out);
+       if (ret) {
+               OSD_DEBUG("_init_blk_request failed\n");
+               return ret;
+       }
+
+       if (or->out.bio) {
+               ret = blk_rq_append_bio(or->request->q, or->out.req,
+                                       or->out.bio);
+               if (ret) {
+                       OSD_DEBUG("blk_rq_append_bio out failed\n");
+                       return ret;
+               }
+               OSD_DEBUG("out bytes=%llu (bytes_req=%u)\n",
+                       _LLU(or->out.total_bytes), or->out.req->data_len);
+       }
+       if (or->in.bio) {
+               ret = blk_rq_append_bio(or->request->q, or->in.req, or->in.bio);
+               if (ret) {
+                       OSD_DEBUG("blk_rq_append_bio in failed\n");
+                       return ret;
+               }
+               OSD_DEBUG("in bytes=%llu (bytes_req=%u)\n",
+                       _LLU(or->in.total_bytes), or->in.req->data_len);
+       }
+
+       if (!or->attributes_mode)
+               or->attributes_mode = OSD_CDB_GET_SET_ATTR_LISTS;
+       cdbh->command_specific_options |= or->attributes_mode;
+
+       or->request->cmd = or->cdb.buff;
+       or->request->cmd_len = _osd_req_cdb_len(or);
+
+       return 0;
+}
+EXPORT_SYMBOL(osd_finalize_request);
+
+/*
+ * Implementation of osd_sec.h API
+ * TODO: Move to a separate osd_sec.c file at a later stage.
+ */
+
+enum { OSD_SEC_CAP_V1_ALL_CAPS =
+       OSD_SEC_CAP_APPEND | OSD_SEC_CAP_OBJ_MGMT | OSD_SEC_CAP_REMOVE   |
+       OSD_SEC_CAP_CREATE | OSD_SEC_CAP_SET_ATTR | OSD_SEC_CAP_GET_ATTR |
+       OSD_SEC_CAP_WRITE  | OSD_SEC_CAP_READ     | OSD_SEC_CAP_POL_SEC  |
+       OSD_SEC_CAP_GLOBAL | OSD_SEC_CAP_DEV_MGMT
+};
+
+void osd_sec_init_nosec_doall_caps(void *caps,
+       const struct osd_obj_id *obj, bool is_collection, const bool is_v1)
+{
+       struct osd_capability *cap = caps;
+       u8 type;
+       u8 descriptor_type;
+
+       if (likely(obj->id)) {
+               if (unlikely(is_collection)) {
+                       type = OSD_SEC_OBJ_COLLECTION;
+                       descriptor_type = is_v1 ? OSD_SEC_OBJ_DESC_OBJ :
+                                                 OSD_SEC_OBJ_DESC_COL;
+               } else {
+                       type = OSD_SEC_OBJ_USER;
+                       descriptor_type = OSD_SEC_OBJ_DESC_OBJ;
+               }
+               WARN_ON(!obj->partition);
+       } else {
+               type = obj->partition ? OSD_SEC_OBJ_PARTITION :
+                                       OSD_SEC_OBJ_ROOT;
+               descriptor_type = OSD_SEC_OBJ_DESC_PAR;
+       }
+
+       memset(cap, 0, sizeof(*cap));
+
+       cap->h.format = OSD_SEC_CAP_FORMAT_VER1;
+       cap->h.integrity_algorithm__key_version = 0; /* MAKE_BYTE(0, 0); */
+       cap->h.security_method = OSD_SEC_NOSEC;
+/*     cap->expiration_time;
+       cap->AUDIT[30-10];
+       cap->discriminator[42-30];
+       cap->object_created_time; */
+       cap->h.object_type = type;
+       osd_sec_set_caps(&cap->h, OSD_SEC_CAP_V1_ALL_CAPS);
+       cap->h.object_descriptor_type = descriptor_type;
+       cap->od.obj_desc.policy_access_tag = 0;
+       cap->od.obj_desc.allowed_partition_id = cpu_to_be64(obj->partition);
+       cap->od.obj_desc.allowed_object_id = cpu_to_be64(obj->id);
+}
+EXPORT_SYMBOL(osd_sec_init_nosec_doall_caps);
+
+void osd_set_caps(struct osd_cdb *cdb, const void *caps)
+{
+       memcpy(&cdb->v1.caps, caps, OSDv1_CAP_LEN);
+}