Merge git://git.kernel.org/pub/scm/linux/kernel/git/jejb/scsi-misc-2.6
[safe/jmp/linux-2.6] / drivers / scsi / iscsi_tcp.c
index fd88777..7608310 100644 (file)
@@ -27,8 +27,8 @@
  */
 
 #include <linux/types.h>
-#include <linux/list.h>
 #include <linux/inet.h>
+#include <linux/slab.h>
 #include <linux/file.h>
 #include <linux/blkdev.h>
 #include <linux/crypto.h>
 
 #include "iscsi_tcp.h"
 
-MODULE_AUTHOR("Dmitry Yusupov <dmitry_yus@yahoo.com>, "
+MODULE_AUTHOR("Mike Christie <michaelc@cs.wisc.edu>, "
+             "Dmitry Yusupov <dmitry_yus@yahoo.com>, "
              "Alex Aizman <itn780@yahoo.com>");
 MODULE_DESCRIPTION("iSCSI/TCP data-path");
 MODULE_LICENSE("GPL");
-#undef DEBUG_TCP
-#define DEBUG_ASSERT
-
-#ifdef DEBUG_TCP
-#define debug_tcp(fmt...) printk(KERN_INFO "tcp: " fmt)
-#else
-#define debug_tcp(fmt...)
-#endif
-
-#ifndef DEBUG_ASSERT
-#ifdef BUG_ON
-#undef BUG_ON
-#endif
-#define BUG_ON(expr)
-#endif
+
+static struct scsi_transport_template *iscsi_sw_tcp_scsi_transport;
+static struct scsi_host_template iscsi_sw_tcp_sht;
+static struct iscsi_transport iscsi_sw_tcp_transport;
 
 static unsigned int iscsi_max_lun = 512;
 module_param_named(max_lun, iscsi_max_lun, uint, S_IRUGO);
 
-static int iscsi_tcp_hdr_recv_done(struct iscsi_tcp_conn *tcp_conn,
-                                  struct iscsi_chunk *chunk);
-
-static inline void
-iscsi_buf_init_iov(struct iscsi_buf *ibuf, char *vbuf, int size)
-{
-       ibuf->sg.page = virt_to_page(vbuf);
-       ibuf->sg.offset = offset_in_page(vbuf);
-       ibuf->sg.length = size;
-       ibuf->sent = 0;
-       ibuf->use_sendmsg = 1;
-}
-
-static inline void
-iscsi_buf_init_sg(struct iscsi_buf *ibuf, struct scatterlist *sg)
-{
-       ibuf->sg.page = sg->page;
-       ibuf->sg.offset = sg->offset;
-       ibuf->sg.length = sg->length;
-       /*
-        * Fastpath: sg element fits into single page
-        */
-       if (sg->length + sg->offset <= PAGE_SIZE && !PageSlab(sg->page))
-               ibuf->use_sendmsg = 0;
-       else
-               ibuf->use_sendmsg = 1;
-       ibuf->sent = 0;
-}
-
-static inline int
-iscsi_buf_left(struct iscsi_buf *ibuf)
-{
-       int rc;
-
-       rc = ibuf->sg.length - ibuf->sent;
-       BUG_ON(rc < 0);
-       return rc;
-}
-
-static inline void
-iscsi_hdr_digest(struct iscsi_conn *conn, struct iscsi_buf *buf,
-                u8* crc)
-{
-       struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
-
-       crypto_hash_digest(&tcp_conn->tx_hash, &buf->sg, buf->sg.length, crc);
-       buf->sg.length += sizeof(u32);
-}
-
-/*
- * Scatterlist handling: inside the iscsi_chunk, we
- * remember an index into the scatterlist, and set data/size
- * to the current scatterlist entry. For highmem pages, we
- * kmap as needed.
- *
- * Note that the page is unmapped when we return from
- * TCP's data_ready handler, so we may end up mapping and
- * unmapping the same page repeatedly. The whole reason
- * for this is that we shouldn't keep the page mapped
- * outside the softirq.
- */
-
-/**
- * iscsi_tcp_chunk_init_sg - init indicated scatterlist entry
- * @chunk: the buffer object
- * @idx: index into scatterlist
- * @offset: byte offset into that sg entry
- *
- * This function sets up the chunk so that subsequent
- * data is copied to the indicated sg entry, at the given
- * offset.
- */
-static inline void
-iscsi_tcp_chunk_init_sg(struct iscsi_chunk *chunk,
-                       unsigned int idx, unsigned int offset)
-{
-       struct scatterlist *sg;
-
-       BUG_ON(chunk->sg == NULL);
-
-       sg = &chunk->sg[idx];
-       chunk->sg_index = idx;
-       chunk->sg_offset = offset;
-       chunk->size = min(sg->length - offset, chunk->total_size);
-       chunk->data = NULL;
-}
-
-/**
- * iscsi_tcp_chunk_map - map the current S/G page
- * @chunk: iscsi chunk
- *
- * We only need to possibly kmap data if scatter lists are being used,
- * because the iscsi passthrough and internal IO paths will never use high
- * mem pages.
- */
-static inline void
-iscsi_tcp_chunk_map(struct iscsi_chunk *chunk)
-{
-       struct scatterlist *sg;
-
-       if (chunk->data != NULL || !chunk->sg)
-               return;
-
-       sg = &chunk->sg[chunk->sg_index];
-       BUG_ON(chunk->sg_mapped);
-       BUG_ON(sg->length == 0);
-       chunk->sg_mapped = kmap_atomic(sg->page, KM_SOFTIRQ0);
-       chunk->data = chunk->sg_mapped + sg->offset + chunk->sg_offset;
-}
-
-static inline void
-iscsi_tcp_chunk_unmap(struct iscsi_chunk *chunk)
-{
-       if (chunk->sg_mapped) {
-               kunmap_atomic(chunk->sg_mapped, KM_SOFTIRQ0);
-               chunk->sg_mapped = NULL;
-               chunk->data = NULL;
-       }
-}
-
-/*
- * Splice the digest buffer into the buffer
- */
-static inline void
-iscsi_tcp_chunk_splice_digest(struct iscsi_chunk *chunk, void *digest)
-{
-       chunk->data = digest;
-       chunk->digest_len = ISCSI_DIGEST_SIZE;
-       chunk->total_size += ISCSI_DIGEST_SIZE;
-       chunk->size = ISCSI_DIGEST_SIZE;
-       chunk->copied = 0;
-       chunk->sg = NULL;
-       chunk->sg_index = 0;
-       chunk->hash = NULL;
-}
-
-/**
- * iscsi_tcp_chunk_done - check whether the chunk is complete
- * @chunk: iscsi chunk to check
- *
- * Check if we're done receiving this chunk. If the receive
- * buffer is full but we expect more data, move on to the
- * next entry in the scatterlist.
- *
- * If the amount of data we received isn't a multiple of 4,
- * we will transparently receive the pad bytes, too.
- *
- * This function must be re-entrant.
- */
-static inline int
-iscsi_tcp_chunk_done(struct iscsi_chunk *chunk)
-{
-       static unsigned char padbuf[ISCSI_PAD_LEN];
-
-       if (chunk->copied < chunk->size) {
-               iscsi_tcp_chunk_map(chunk);
-               return 0;
-       }
-
-       chunk->total_copied += chunk->copied;
-       chunk->copied = 0;
-       chunk->size = 0;
-
-       /* Unmap the current scatterlist page, if there is one. */
-       iscsi_tcp_chunk_unmap(chunk);
-
-       /* Do we have more scatterlist entries? */
-       if (chunk->total_copied < chunk->total_size) {
-               /* Proceed to the next entry in the scatterlist. */
-               iscsi_tcp_chunk_init_sg(chunk, chunk->sg_index + 1, 0);
-               iscsi_tcp_chunk_map(chunk);
-               BUG_ON(chunk->size == 0);
-               return 0;
-       }
-
-       /* Do we need to handle padding? */
-       if (chunk->total_copied & (ISCSI_PAD_LEN-1)) {
-               unsigned int pad;
-
-               pad = ISCSI_PAD_LEN - (chunk->total_copied & (ISCSI_PAD_LEN-1));
-               debug_tcp("consume %d pad bytes\n", pad);
-               chunk->total_size += pad;
-               chunk->size = pad;
-               chunk->data = padbuf;
-               return 0;
-       }
-
-       /*
-        * Set us up for receiving the data digest. hdr digest
-        * is completely handled in hdr done function.
-        */
-       if (chunk->hash) {
-               if (chunk->digest_len == 0) {
-                       crypto_hash_final(chunk->hash, chunk->digest);
-                       iscsi_tcp_chunk_splice_digest(chunk,
-                                                     chunk->recv_digest);
-                       return 0;
-               }
-       }
-
-       return 1;
-}
-
-/**
- * iscsi_tcp_chunk_recv - copy data to chunk
- * @tcp_conn: the iSCSI TCP connection
- * @chunk: the buffer to copy to
- * @ptr: data pointer
- * @len: amount of data available
- *
- * This function copies up to @len bytes to the
- * given buffer, and returns the number of bytes
- * consumed, which can actually be less than @len.
- *
- * If hash digest is enabled, the function will update the
- * hash while copying.
- * Combining these two operations doesn't buy us a lot (yet),
- * but in the future we could implement combined copy+crc,
- * just way we do for network layer checksums.
- */
-static int
-iscsi_tcp_chunk_recv(struct iscsi_tcp_conn *tcp_conn,
-                    struct iscsi_chunk *chunk, const void *ptr,
-                    unsigned int len)
-{
-       struct scatterlist sg;
-       unsigned int copy, copied = 0;
-
-       while (!iscsi_tcp_chunk_done(chunk)) {
-               if (copied == len)
-                       goto out;
-
-               copy = min(len - copied, chunk->size - chunk->copied);
-               memcpy(chunk->data + chunk->copied, ptr + copied, copy);
-
-               if (chunk->hash) {
-                       sg_init_one(&sg, ptr + copied, copy);
-                       crypto_hash_update(chunk->hash, &sg, copy);
-               }
-               chunk->copied += copy;
-               copied += copy;
-       }
-
-out:
-       return copied;
-}
-
-static inline void
-iscsi_tcp_dgst_header(struct hash_desc *hash, const void *hdr, size_t hdrlen,
-                     unsigned char digest[ISCSI_DIGEST_SIZE])
-{
-       struct scatterlist sg;
-
-       sg_init_one(&sg, hdr, hdrlen);
-       crypto_hash_digest(hash, &sg, hdrlen, digest);
-}
-
-static inline int
-iscsi_tcp_dgst_verify(struct iscsi_tcp_conn *tcp_conn,
-                     struct iscsi_chunk *chunk)
-{
-       if (!chunk->digest_len)
-               return 1;
-
-       if (memcmp(chunk->recv_digest, chunk->digest, chunk->digest_len)) {
-               debug_scsi("digest mismatch\n");
-               return 0;
-       }
-
-       return 1;
-}
-
-/*
- * Helper function to set up chunk buffer
- */
-static inline void
-__iscsi_chunk_init(struct iscsi_chunk *chunk, size_t size,
-                  iscsi_chunk_done_fn_t *done, struct hash_desc *hash)
-{
-       memset(chunk, 0, sizeof(*chunk));
-       chunk->total_size = size;
-       chunk->done = done;
-
-       if (hash) {
-               chunk->hash = hash;
-               crypto_hash_init(hash);
-       }
-}
-
-static inline void
-iscsi_chunk_init_linear(struct iscsi_chunk *chunk, void *data, size_t size,
-                       iscsi_chunk_done_fn_t *done, struct hash_desc *hash)
-{
-       __iscsi_chunk_init(chunk, size, done, hash);
-       chunk->data = data;
-       chunk->size = size;
-}
-
-static inline int
-iscsi_chunk_seek_sg(struct iscsi_chunk *chunk,
-                   struct scatterlist *sg, unsigned int sg_count,
-                   unsigned int offset, size_t size,
-                   iscsi_chunk_done_fn_t *done, struct hash_desc *hash)
-{
-       unsigned int i;
-
-       __iscsi_chunk_init(chunk, size, done, hash);
-       for (i = 0; i < sg_count; ++i) {
-               if (offset < sg[i].length) {
-                       chunk->sg = sg;
-                       chunk->sg_count = sg_count;
-                       iscsi_tcp_chunk_init_sg(chunk, i, offset);
-                       return 0;
-               }
-               offset -= sg[i].length;
-       }
-
-       return ISCSI_ERR_DATA_OFFSET;
-}
-
-/**
- * iscsi_tcp_hdr_recv_prep - prep chunk for hdr reception
- * @tcp_conn: iscsi connection to prep for
- *
- * This function always passes NULL for the hash argument, because when this
- * function is called we do not yet know the final size of the header and want
- * to delay the digest processing until we know that.
- */
-static void
-iscsi_tcp_hdr_recv_prep(struct iscsi_tcp_conn *tcp_conn)
-{
-       debug_tcp("iscsi_tcp_hdr_recv_prep(%p%s)\n", tcp_conn,
-                 tcp_conn->iscsi_conn->hdrdgst_en ? ", digest enabled" : "");
-       iscsi_chunk_init_linear(&tcp_conn->in.chunk,
-                               tcp_conn->in.hdr_buf, sizeof(struct iscsi_hdr),
-                               iscsi_tcp_hdr_recv_done, NULL);
-}
-
-/*
- * Handle incoming reply to any other type of command
- */
-static int
-iscsi_tcp_data_recv_done(struct iscsi_tcp_conn *tcp_conn,
-                        struct iscsi_chunk *chunk)
-{
-       struct iscsi_conn *conn = tcp_conn->iscsi_conn;
-       int rc = 0;
-
-       if (!iscsi_tcp_dgst_verify(tcp_conn, chunk))
-               return ISCSI_ERR_DATA_DGST;
-
-       rc = iscsi_complete_pdu(conn, tcp_conn->in.hdr,
-                       conn->data, tcp_conn->in.datalen);
-       if (rc)
-               return rc;
-
-       iscsi_tcp_hdr_recv_prep(tcp_conn);
-       return 0;
-}
-
-static void
-iscsi_tcp_data_recv_prep(struct iscsi_tcp_conn *tcp_conn)
-{
-       struct iscsi_conn *conn = tcp_conn->iscsi_conn;
-       struct hash_desc *rx_hash = NULL;
-
-       if (conn->datadgst_en)
-               rx_hash = &tcp_conn->rx_hash;
-
-       iscsi_chunk_init_linear(&tcp_conn->in.chunk,
-                               conn->data, tcp_conn->in.datalen,
-                               iscsi_tcp_data_recv_done, rx_hash);
-}
-
-/*
- * must be called with session lock
- */
-static void
-iscsi_tcp_cleanup_ctask(struct iscsi_conn *conn, struct iscsi_cmd_task *ctask)
-{
-       struct iscsi_tcp_cmd_task *tcp_ctask = ctask->dd_data;
-       struct iscsi_r2t_info *r2t;
-       struct scsi_cmnd *sc;
-
-       /* flush ctask's r2t queues */
-       while (__kfifo_get(tcp_ctask->r2tqueue, (void*)&r2t, sizeof(void*))) {
-               __kfifo_put(tcp_ctask->r2tpool.queue, (void*)&r2t,
-                           sizeof(void*));
-               debug_scsi("iscsi_tcp_cleanup_ctask pending r2t dropped\n");
-       }
-
-       sc = ctask->sc;
-       if (unlikely(!sc))
-               return;
-
-       tcp_ctask->xmstate = XMSTATE_IDLE;
-       tcp_ctask->r2t = NULL;
-}
-
-/**
- * iscsi_data_rsp - SCSI Data-In Response processing
- * @conn: iscsi connection
- * @ctask: scsi command task
- **/
-static int
-iscsi_data_rsp(struct iscsi_conn *conn, struct iscsi_cmd_task *ctask)
-{
-       struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
-       struct iscsi_tcp_cmd_task *tcp_ctask = ctask->dd_data;
-       struct iscsi_data_rsp *rhdr = (struct iscsi_data_rsp *)tcp_conn->in.hdr;
-       struct iscsi_session *session = conn->session;
-       struct scsi_cmnd *sc = ctask->sc;
-       int datasn = be32_to_cpu(rhdr->datasn);
-
-       iscsi_update_cmdsn(session, (struct iscsi_nopin*)rhdr);
-       /*
-        * setup Data-In byte counter (gets decremented..)
-        */
-       ctask->data_count = tcp_conn->in.datalen;
-
-       if (tcp_conn->in.datalen == 0)
-               return 0;
-
-       if (tcp_ctask->exp_datasn != datasn) {
-               debug_tcp("%s: ctask->exp_datasn(%d) != rhdr->datasn(%d)\n",
-                         __FUNCTION__, tcp_ctask->exp_datasn, datasn);
-               return ISCSI_ERR_DATASN;
-       }
-
-       tcp_ctask->exp_datasn++;
-
-       tcp_ctask->data_offset = be32_to_cpu(rhdr->offset);
-       if (tcp_ctask->data_offset + tcp_conn->in.datalen > scsi_bufflen(sc)) {
-               debug_tcp("%s: data_offset(%d) + data_len(%d) > total_length_in(%d)\n",
-                         __FUNCTION__, tcp_ctask->data_offset,
-                         tcp_conn->in.datalen, scsi_bufflen(sc));
-               return ISCSI_ERR_DATA_OFFSET;
-       }
-
-       if (rhdr->flags & ISCSI_FLAG_DATA_STATUS) {
-               sc->result = (DID_OK << 16) | rhdr->cmd_status;
-               conn->exp_statsn = be32_to_cpu(rhdr->statsn) + 1;
-               if (rhdr->flags & (ISCSI_FLAG_DATA_UNDERFLOW |
-                                  ISCSI_FLAG_DATA_OVERFLOW)) {
-                       int res_count = be32_to_cpu(rhdr->residual_count);
-
-                       if (res_count > 0 &&
-                           (rhdr->flags & ISCSI_FLAG_CMD_OVERFLOW ||
-                            res_count <= scsi_bufflen(sc)))
-                               scsi_set_resid(sc, res_count);
-                       else
-                               sc->result = (DID_BAD_TARGET << 16) |
-                                       rhdr->cmd_status;
-               }
-       }
-
-       conn->datain_pdus_cnt++;
-       return 0;
-}
-
-/**
- * iscsi_solicit_data_init - initialize first Data-Out
- * @conn: iscsi connection
- * @ctask: scsi command task
- * @r2t: R2T info
- *
- * Notes:
- *     Initialize first Data-Out within this R2T sequence and finds
- *     proper data_offset within this SCSI command.
- *
- *     This function is called with connection lock taken.
- **/
-static void
-iscsi_solicit_data_init(struct iscsi_conn *conn, struct iscsi_cmd_task *ctask,
-                       struct iscsi_r2t_info *r2t)
-{
-       struct iscsi_data *hdr;
-       struct scsi_cmnd *sc = ctask->sc;
-       int i, sg_count = 0;
-       struct scatterlist *sg;
-
-       hdr = &r2t->dtask.hdr;
-       memset(hdr, 0, sizeof(struct iscsi_data));
-       hdr->ttt = r2t->ttt;
-       hdr->datasn = cpu_to_be32(r2t->solicit_datasn);
-       r2t->solicit_datasn++;
-       hdr->opcode = ISCSI_OP_SCSI_DATA_OUT;
-       memcpy(hdr->lun, ctask->hdr->lun, sizeof(hdr->lun));
-       hdr->itt = ctask->hdr->itt;
-       hdr->exp_statsn = r2t->exp_statsn;
-       hdr->offset = cpu_to_be32(r2t->data_offset);
-       if (r2t->data_length > conn->max_xmit_dlength) {
-               hton24(hdr->dlength, conn->max_xmit_dlength);
-               r2t->data_count = conn->max_xmit_dlength;
-               hdr->flags = 0;
-       } else {
-               hton24(hdr->dlength, r2t->data_length);
-               r2t->data_count = r2t->data_length;
-               hdr->flags = ISCSI_FLAG_CMD_FINAL;
-       }
-       conn->dataout_pdus_cnt++;
-
-       r2t->sent = 0;
-
-       iscsi_buf_init_iov(&r2t->headbuf, (char*)hdr,
-                          sizeof(struct iscsi_hdr));
-
-       sg = scsi_sglist(sc);
-       r2t->sg = NULL;
-       for (i = 0; i < scsi_sg_count(sc); i++, sg += 1) {
-               /* FIXME: prefetch ? */
-               if (sg_count + sg->length > r2t->data_offset) {
-                       int page_offset;
-
-                       /* sg page found! */
-
-                       /* offset within this page */
-                       page_offset = r2t->data_offset - sg_count;
-
-                       /* fill in this buffer */
-                       iscsi_buf_init_sg(&r2t->sendbuf, sg);
-                       r2t->sendbuf.sg.offset += page_offset;
-                       r2t->sendbuf.sg.length -= page_offset;
-
-                       /* xmit logic will continue with next one */
-                       r2t->sg = sg + 1;
-                       break;
-               }
-               sg_count += sg->length;
-       }
-       BUG_ON(r2t->sg == NULL);
-}
-
-/**
- * iscsi_r2t_rsp - iSCSI R2T Response processing
- * @conn: iscsi connection
- * @ctask: scsi command task
- **/
-static int
-iscsi_r2t_rsp(struct iscsi_conn *conn, struct iscsi_cmd_task *ctask)
-{
-       struct iscsi_r2t_info *r2t;
-       struct iscsi_session *session = conn->session;
-       struct iscsi_tcp_cmd_task *tcp_ctask = ctask->dd_data;
-       struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
-       struct iscsi_r2t_rsp *rhdr = (struct iscsi_r2t_rsp *)tcp_conn->in.hdr;
-       int r2tsn = be32_to_cpu(rhdr->r2tsn);
-       int rc;
-
-       if (tcp_conn->in.datalen) {
-               printk(KERN_ERR "iscsi_tcp: invalid R2t with datalen %d\n",
-                      tcp_conn->in.datalen);
-               return ISCSI_ERR_DATALEN;
-       }
-
-       if (tcp_ctask->exp_datasn != r2tsn){
-               debug_tcp("%s: ctask->exp_datasn(%d) != rhdr->r2tsn(%d)\n",
-                         __FUNCTION__, tcp_ctask->exp_datasn, r2tsn);
-               return ISCSI_ERR_R2TSN;
-       }
-
-       /* fill-in new R2T associated with the task */
-       spin_lock(&session->lock);
-       iscsi_update_cmdsn(session, (struct iscsi_nopin*)rhdr);
-
-       if (!ctask->sc || session->state != ISCSI_STATE_LOGGED_IN) {
-               printk(KERN_INFO "iscsi_tcp: dropping R2T itt %d in "
-                      "recovery...\n", ctask->itt);
-               spin_unlock(&session->lock);
-               return 0;
-       }
-
-       rc = __kfifo_get(tcp_ctask->r2tpool.queue, (void*)&r2t, sizeof(void*));
-       BUG_ON(!rc);
-
-       r2t->exp_statsn = rhdr->statsn;
-       r2t->data_length = be32_to_cpu(rhdr->data_length);
-       if (r2t->data_length == 0) {
-               printk(KERN_ERR "iscsi_tcp: invalid R2T with zero data len\n");
-               spin_unlock(&session->lock);
-               return ISCSI_ERR_DATALEN;
-       }
-
-       if (r2t->data_length > session->max_burst)
-               debug_scsi("invalid R2T with data len %u and max burst %u."
-                          "Attempting to execute request.\n",
-                           r2t->data_length, session->max_burst);
-
-       r2t->data_offset = be32_to_cpu(rhdr->data_offset);
-       if (r2t->data_offset + r2t->data_length > scsi_bufflen(ctask->sc)) {
-               spin_unlock(&session->lock);
-               printk(KERN_ERR "iscsi_tcp: invalid R2T with data len %u at "
-                      "offset %u and total length %d\n", r2t->data_length,
-                      r2t->data_offset, scsi_bufflen(ctask->sc));
-               return ISCSI_ERR_DATALEN;
-       }
-
-       r2t->ttt = rhdr->ttt; /* no flip */
-       r2t->solicit_datasn = 0;
-
-       iscsi_solicit_data_init(conn, ctask, r2t);
-
-       tcp_ctask->exp_datasn = r2tsn + 1;
-       __kfifo_put(tcp_ctask->r2tqueue, (void*)&r2t, sizeof(void*));
-       tcp_ctask->xmstate |= XMSTATE_SOL_HDR_INIT;
-       conn->r2t_pdus_cnt++;
-
-       iscsi_requeue_ctask(ctask);
-       spin_unlock(&session->lock);
+static int iscsi_sw_tcp_dbg;
+module_param_named(debug_iscsi_tcp, iscsi_sw_tcp_dbg, int,
+                  S_IRUGO | S_IWUSR);
+MODULE_PARM_DESC(debug_iscsi_tcp, "Turn on debugging for iscsi_tcp module "
+                "Set to 1 to turn on, and zero to turn off. Default is off.");
 
-       return 0;
-}
+#define ISCSI_SW_TCP_DBG(_conn, dbg_fmt, arg...)               \
+       do {                                                    \
+               if (iscsi_sw_tcp_dbg)                           \
+                       iscsi_conn_printk(KERN_INFO, _conn,     \
+                                            "%s " dbg_fmt,     \
+                                            __func__, ##arg);  \
+       } while (0);
 
-/*
- * Handle incoming reply to DataIn command
- */
-static int
-iscsi_tcp_process_data_in(struct iscsi_tcp_conn *tcp_conn,
-                         struct iscsi_chunk *chunk)
-{
-       struct iscsi_conn *conn = tcp_conn->iscsi_conn;
-       struct iscsi_hdr *hdr = tcp_conn->in.hdr;
-       int rc;
-
-       if (!iscsi_tcp_dgst_verify(tcp_conn, chunk))
-               return ISCSI_ERR_DATA_DGST;
-
-       /* check for non-exceptional status */
-       if (hdr->flags & ISCSI_FLAG_DATA_STATUS) {
-               rc = iscsi_complete_pdu(conn, tcp_conn->in.hdr, NULL, 0);
-               if (rc)
-                       return rc;
-       }
-
-       iscsi_tcp_hdr_recv_prep(tcp_conn);
-       return 0;
-}
 
 /**
- * iscsi_tcp_hdr_dissect - process PDU header
- * @conn: iSCSI connection
- * @hdr: PDU header
- *
- * This function analyzes the header of the PDU received,
- * and performs several sanity checks. If the PDU is accompanied
- * by data, the receive buffer is set up to copy the incoming data
- * to the correct location.
+ * iscsi_sw_tcp_recv - TCP receive in sendfile fashion
+ * @rd_desc: read descriptor
+ * @skb: socket buffer
+ * @offset: offset in skb
+ * @len: skb->len - offset
  */
-static int
-iscsi_tcp_hdr_dissect(struct iscsi_conn *conn, struct iscsi_hdr *hdr)
+static int iscsi_sw_tcp_recv(read_descriptor_t *rd_desc, struct sk_buff *skb,
+                            unsigned int offset, size_t len)
 {
-       int rc = 0, opcode, ahslen;
-       struct iscsi_session *session = conn->session;
-       struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
-       struct iscsi_cmd_task *ctask;
-       uint32_t itt;
-
-       /* verify PDU length */
-       tcp_conn->in.datalen = ntoh24(hdr->dlength);
-       if (tcp_conn->in.datalen > conn->max_recv_dlength) {
-               printk(KERN_ERR "iscsi_tcp: datalen %d > %d\n",
-                      tcp_conn->in.datalen, conn->max_recv_dlength);
-               return ISCSI_ERR_DATALEN;
-       }
-
-       /* Additional header segments. So far, we don't
-        * process additional headers.
-        */
-       ahslen = hdr->hlength << 2;
-
-       opcode = hdr->opcode & ISCSI_OPCODE_MASK;
-       /* verify itt (itt encoding: age+cid+itt) */
-       rc = iscsi_verify_itt(conn, hdr, &itt);
-       if (rc == ISCSI_ERR_NO_SCSI_CMD) {
-               /* XXX: what does this do? */
-               tcp_conn->in.datalen = 0; /* force drop */
-               return 0;
-       } else if (rc)
-               return rc;
-
-       debug_tcp("opcode 0x%x ahslen %d datalen %d\n",
-                 opcode, ahslen, tcp_conn->in.datalen);
-
-       switch(opcode) {
-       case ISCSI_OP_SCSI_DATA_IN:
-               ctask = session->cmds[itt];
-               rc = iscsi_data_rsp(conn, ctask);
-               if (rc)
-                       return rc;
-               if (tcp_conn->in.datalen) {
-                       struct iscsi_tcp_cmd_task *tcp_ctask = ctask->dd_data;
-                       struct hash_desc *rx_hash = NULL;
-
-                       /*
-                        * Setup copy of Data-In into the Scsi_Cmnd
-                        * Scatterlist case:
-                        * We set up the iscsi_chunk to point to the next
-                        * scatterlist entry to copy to. As we go along,
-                        * we move on to the next scatterlist entry and
-                        * update the digest per-entry.
-                        */
-                       if (conn->datadgst_en)
-                               rx_hash = &tcp_conn->rx_hash;
-
-                       debug_tcp("iscsi_tcp_begin_data_in(%p, offset=%d, "
-                                 "datalen=%d)\n", tcp_conn,
-                                 tcp_ctask->data_offset,
-                                 tcp_conn->in.datalen);
-                       return iscsi_chunk_seek_sg(&tcp_conn->in.chunk,
-                                               scsi_sglist(ctask->sc),
-                                               scsi_sg_count(ctask->sc),
-                                               tcp_ctask->data_offset,
-                                               tcp_conn->in.datalen,
-                                               iscsi_tcp_process_data_in,
-                                               rx_hash);
-               }
-               /* fall through */
-       case ISCSI_OP_SCSI_CMD_RSP:
-               if (tcp_conn->in.datalen) {
-                       iscsi_tcp_data_recv_prep(tcp_conn);
-                       return 0;
-               }
-               rc = iscsi_complete_pdu(conn, hdr, NULL, 0);
-               break;
-       case ISCSI_OP_R2T:
-               ctask = session->cmds[itt];
-               if (ahslen)
-                       rc = ISCSI_ERR_AHSLEN;
-               else if (ctask->sc->sc_data_direction == DMA_TO_DEVICE)
-                       rc = iscsi_r2t_rsp(conn, ctask);
-               else
-                       rc = ISCSI_ERR_PROTO;
-               break;
-       case ISCSI_OP_LOGIN_RSP:
-       case ISCSI_OP_TEXT_RSP:
-       case ISCSI_OP_REJECT:
-       case ISCSI_OP_ASYNC_EVENT:
-               /*
-                * It is possible that we could get a PDU with a buffer larger
-                * than 8K, but there are no targets that currently do this.
-                * For now we fail until we find a vendor that needs it
-                */
-               if (ISCSI_DEF_MAX_RECV_SEG_LEN < tcp_conn->in.datalen) {
-                       printk(KERN_ERR "iscsi_tcp: received buffer of len %u "
-                             "but conn buffer is only %u (opcode %0x)\n",
-                             tcp_conn->in.datalen,
-                             ISCSI_DEF_MAX_RECV_SEG_LEN, opcode);
-                       rc = ISCSI_ERR_PROTO;
-                       break;
-               }
+       struct iscsi_conn *conn = rd_desc->arg.data;
+       unsigned int consumed, total_consumed = 0;
+       int status;
 
-               /* If there's data coming in with the response,
-                * receive it to the connection's buffer.
-                */
-               if (tcp_conn->in.datalen) {
-                       iscsi_tcp_data_recv_prep(tcp_conn);
-                       return 0;
-               }
-       /* fall through */
-       case ISCSI_OP_LOGOUT_RSP:
-       case ISCSI_OP_NOOP_IN:
-       case ISCSI_OP_SCSI_TMFUNC_RSP:
-               rc = iscsi_complete_pdu(conn, hdr, NULL, 0);
-               break;
-       default:
-               rc = ISCSI_ERR_BAD_OPCODE;
-               break;
-       }
+       ISCSI_SW_TCP_DBG(conn, "in %d bytes\n", skb->len - offset);
 
-       if (rc == 0) {
-               /* Anything that comes with data should have
-                * been handled above. */
-               if (tcp_conn->in.datalen)
-                       return ISCSI_ERR_PROTO;
-               iscsi_tcp_hdr_recv_prep(tcp_conn);
-       }
+       do {
+               status = 0;
+               consumed = iscsi_tcp_recv_skb(conn, skb, offset, 0, &status);
+               offset += consumed;
+               total_consumed += consumed;
+       } while (consumed != 0 && status != ISCSI_TCP_SKB_DONE);
 
-       return rc;
-}
-
-static inline void
-partial_sg_digest_update(struct hash_desc *desc, struct scatterlist *sg,
-                        int offset, int length)
-{
-       struct scatterlist temp;
-
-       sg_init_table(&temp, 1);
-       sg_set_page(&temp, sg_page(sg), length, offset);
-       crypto_hash_update(desc, &temp, length);
+       ISCSI_SW_TCP_DBG(conn, "read %d bytes status %d\n",
+                        skb->len - offset, status);
+       return total_consumed;
 }
 
 /**
- * iscsi_tcp_hdr_recv_done - process PDU header
+ * iscsi_sw_sk_state_check - check socket state
+ * @sk: socket
  *
- * This is the callback invoked when the PDU header has
- * been received. If the header is followed by additional
- * header segments, we go back for more data.
+ * If the socket is in CLOSE or CLOSE_WAIT we should
+ * not close the connection if there is still some
+ * data pending.
  */
-static int
-iscsi_tcp_hdr_recv_done(struct iscsi_tcp_conn *tcp_conn,
-                       struct iscsi_chunk *chunk)
-{
-       struct iscsi_conn *conn = tcp_conn->iscsi_conn;
-       struct iscsi_hdr *hdr;
-
-       /* Check if there are additional header segments
-        * *prior* to computing the digest, because we
-        * may need to go back to the caller for more.
-        */
-       hdr = (struct iscsi_hdr *) tcp_conn->in.hdr_buf;
-       if (chunk->copied == sizeof(struct iscsi_hdr) && hdr->hlength) {
-               /* Bump the header length - the caller will
-                * just loop around and get the AHS for us, and
-                * call again. */
-               unsigned int ahslen = hdr->hlength << 2;
-
-               /* Make sure we don't overflow */
-               if (sizeof(*hdr) + ahslen > sizeof(tcp_conn->in.hdr_buf))
-                       return ISCSI_ERR_AHSLEN;
-
-               chunk->total_size += ahslen;
-               chunk->size += ahslen;
-               return 0;
-       }
-
-       /* We're done processing the header. See if we're doing
-        * header digests; if so, set up the recv_digest buffer
-        * and go back for more. */
-       if (conn->hdrdgst_en) {
-               if (chunk->digest_len == 0) {
-                       iscsi_tcp_chunk_splice_digest(chunk,
-                                                     chunk->recv_digest);
-                       return 0;
-               }
-               iscsi_tcp_dgst_header(&tcp_conn->rx_hash, hdr,
-                                     chunk->total_copied - ISCSI_DIGEST_SIZE,
-                                     chunk->digest);
-
-               if (!iscsi_tcp_dgst_verify(tcp_conn, chunk))
-                       return ISCSI_ERR_HDR_DGST;
-       }
-
-       tcp_conn->in.hdr = hdr;
-       return iscsi_tcp_hdr_dissect(conn, hdr);
-}
-
-/**
- * iscsi_tcp_recv - TCP receive in sendfile fashion
- * @rd_desc: read descriptor
- * @skb: socket buffer
- * @offset: offset in skb
- * @len: skb->len - offset
- **/
-static int
-iscsi_tcp_recv(read_descriptor_t *rd_desc, struct sk_buff *skb,
-              unsigned int offset, size_t len)
+static inline int iscsi_sw_sk_state_check(struct sock *sk)
 {
-       struct iscsi_conn *conn = rd_desc->arg.data;
-       struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
-       struct iscsi_chunk *chunk = &tcp_conn->in.chunk;
-       struct skb_seq_state seq;
-       unsigned int consumed = 0;
-       int rc = 0;
-
-       debug_tcp("in %d bytes\n", skb->len - offset);
-
-       if (unlikely(conn->suspend_rx)) {
-               debug_tcp("conn %d Rx suspended!\n", conn->id);
-               return 0;
-       }
-
-       skb_prepare_seq_read(skb, offset, skb->len, &seq);
-       while (1) {
-               unsigned int avail;
-               const u8 *ptr;
-
-               avail = skb_seq_read(consumed, &ptr, &seq);
-               if (avail == 0)
-                       break;
-               BUG_ON(chunk->copied >= chunk->size);
-
-               debug_tcp("skb %p ptr=%p avail=%u\n", skb, ptr, avail);
-               rc = iscsi_tcp_chunk_recv(tcp_conn, chunk, ptr, avail);
-               BUG_ON(rc == 0);
-               consumed += rc;
-
-               if (chunk->total_copied >= chunk->total_size) {
-                       rc = chunk->done(tcp_conn, chunk);
-                       if (rc != 0) {
-                               skb_abort_seq_read(&seq);
-                               goto error;
-                       }
+       struct iscsi_conn *conn = (struct iscsi_conn*)sk->sk_user_data;
 
-                       /* The done() functions sets up the
-                        * next chunk. */
-               }
+       if ((sk->sk_state == TCP_CLOSE_WAIT || sk->sk_state == TCP_CLOSE) &&
+           !atomic_read(&sk->sk_rmem_alloc)) {
+               ISCSI_SW_TCP_DBG(conn, "TCP_CLOSE|TCP_CLOSE_WAIT\n");
+               iscsi_conn_failure(conn, ISCSI_ERR_TCP_CONN_CLOSE);
+               return -ECONNRESET;
        }
-
-       conn->rxdata_octets += consumed;
-       return consumed;
-
-error:
-       debug_tcp("Error receiving PDU, errno=%d\n", rc);
-       iscsi_conn_failure(conn, ISCSI_ERR_CONN_FAILED);
        return 0;
 }
 
-static void
-iscsi_tcp_data_ready(struct sock *sk, int flag)
+static void iscsi_sw_tcp_data_ready(struct sock *sk, int flag)
 {
        struct iscsi_conn *conn = sk->sk_user_data;
        struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
@@ -991,885 +137,484 @@ iscsi_tcp_data_ready(struct sock *sk, int flag)
         */
        rd_desc.arg.data = conn;
        rd_desc.count = 1;
-       tcp_read_sock(sk, &rd_desc, iscsi_tcp_recv);
+       tcp_read_sock(sk, &rd_desc, iscsi_sw_tcp_recv);
+
+       iscsi_sw_sk_state_check(sk);
 
        read_unlock(&sk->sk_callback_lock);
 
        /* If we had to (atomically) map a highmem page,
         * unmap it now. */
-       iscsi_tcp_chunk_unmap(&tcp_conn->in.chunk);
+       iscsi_tcp_segment_unmap(&tcp_conn->in.segment);
 }
 
-static void
-iscsi_tcp_state_change(struct sock *sk)
+static void iscsi_sw_tcp_state_change(struct sock *sk)
 {
        struct iscsi_tcp_conn *tcp_conn;
+       struct iscsi_sw_tcp_conn *tcp_sw_conn;
        struct iscsi_conn *conn;
        struct iscsi_session *session;
        void (*old_state_change)(struct sock *);
 
-       read_lock(&sk->sk_callback_lock);
-
-       conn = (struct iscsi_conn*)sk->sk_user_data;
-       session = conn->session;
-
-       if ((sk->sk_state == TCP_CLOSE_WAIT ||
-            sk->sk_state == TCP_CLOSE) &&
-           !atomic_read(&sk->sk_rmem_alloc)) {
-               debug_tcp("iscsi_tcp_state_change: TCP_CLOSE|TCP_CLOSE_WAIT\n");
-               iscsi_conn_failure(conn, ISCSI_ERR_CONN_FAILED);
-       }
-
-       tcp_conn = conn->dd_data;
-       old_state_change = tcp_conn->old_state_change;
-
-       read_unlock(&sk->sk_callback_lock);
-
-       old_state_change(sk);
-}
-
-/**
- * iscsi_write_space - Called when more output buffer space is available
- * @sk: socket space is available for
- **/
-static void
-iscsi_write_space(struct sock *sk)
-{
-       struct iscsi_conn *conn = (struct iscsi_conn*)sk->sk_user_data;
-       struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
-
-       tcp_conn->old_write_space(sk);
-       debug_tcp("iscsi_write_space: cid %d\n", conn->id);
-       scsi_queue_work(conn->session->host, &conn->xmitwork);
-}
-
-static void
-iscsi_conn_set_callbacks(struct iscsi_conn *conn)
-{
-       struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
-       struct sock *sk = tcp_conn->sock->sk;
-
-       /* assign new callbacks */
-       write_lock_bh(&sk->sk_callback_lock);
-       sk->sk_user_data = conn;
-       tcp_conn->old_data_ready = sk->sk_data_ready;
-       tcp_conn->old_state_change = sk->sk_state_change;
-       tcp_conn->old_write_space = sk->sk_write_space;
-       sk->sk_data_ready = iscsi_tcp_data_ready;
-       sk->sk_state_change = iscsi_tcp_state_change;
-       sk->sk_write_space = iscsi_write_space;
-       write_unlock_bh(&sk->sk_callback_lock);
-}
-
-static void
-iscsi_conn_restore_callbacks(struct iscsi_tcp_conn *tcp_conn)
-{
-       struct sock *sk = tcp_conn->sock->sk;
-
-       /* restore socket callbacks, see also: iscsi_conn_set_callbacks() */
-       write_lock_bh(&sk->sk_callback_lock);
-       sk->sk_user_data    = NULL;
-       sk->sk_data_ready   = tcp_conn->old_data_ready;
-       sk->sk_state_change = tcp_conn->old_state_change;
-       sk->sk_write_space  = tcp_conn->old_write_space;
-       sk->sk_no_check  = 0;
-       write_unlock_bh(&sk->sk_callback_lock);
-}
-
-/**
- * iscsi_send - generic send routine
- * @sk: kernel's socket
- * @buf: buffer to write from
- * @size: actual size to write
- * @flags: socket's flags
- */
-static inline int
-iscsi_send(struct iscsi_conn *conn, struct iscsi_buf *buf, int size, int flags)
-{
-       struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
-       struct socket *sk = tcp_conn->sock;
-       int offset = buf->sg.offset + buf->sent, res;
-
-       /*
-        * if we got use_sg=0 or are sending something we kmallocd
-        * then we did not have to do kmap (kmap returns page_address)
-        *
-        * if we got use_sg > 0, but had to drop down, we do not
-        * set clustering so this should only happen for that
-        * slab case.
-        */
-       if (buf->use_sendmsg)
-               res = sock_no_sendpage(sk, buf->sg.page, offset, size, flags);
-       else
-               res = tcp_conn->sendpage(sk, buf->sg.page, offset, size, flags);
-
-       if (res >= 0) {
-               conn->txdata_octets += res;
-               buf->sent += res;
-               return res;
-       }
-
-       tcp_conn->sendpage_failures_cnt++;
-       if (res == -EAGAIN)
-               res = -ENOBUFS;
-       else
-               iscsi_conn_failure(conn, ISCSI_ERR_CONN_FAILED);
-       return res;
-}
-
-/**
- * iscsi_sendhdr - send PDU Header via tcp_sendpage()
- * @conn: iscsi connection
- * @buf: buffer to write from
- * @datalen: lenght of data to be sent after the header
- *
- * Notes:
- *     (Tx, Fast Path)
- **/
-static inline int
-iscsi_sendhdr(struct iscsi_conn *conn, struct iscsi_buf *buf, int datalen)
-{
-       int flags = 0; /* MSG_DONTWAIT; */
-       int res, size;
-
-       size = buf->sg.length - buf->sent;
-       BUG_ON(buf->sent + size > buf->sg.length);
-       if (buf->sent + size != buf->sg.length || datalen)
-               flags |= MSG_MORE;
-
-       res = iscsi_send(conn, buf, size, flags);
-       debug_tcp("sendhdr %d bytes, sent %d res %d\n", size, buf->sent, res);
-       if (res >= 0) {
-               if (size != res)
-                       return -EAGAIN;
-               return 0;
-       }
-
-       return res;
-}
-
-/**
- * iscsi_sendpage - send one page of iSCSI Data-Out.
- * @conn: iscsi connection
- * @buf: buffer to write from
- * @count: remaining data
- * @sent: number of bytes sent
- *
- * Notes:
- *     (Tx, Fast Path)
- **/
-static inline int
-iscsi_sendpage(struct iscsi_conn *conn, struct iscsi_buf *buf,
-              int *count, int *sent)
-{
-       int flags = 0; /* MSG_DONTWAIT; */
-       int res, size;
-
-       size = buf->sg.length - buf->sent;
-       BUG_ON(buf->sent + size > buf->sg.length);
-       if (size > *count)
-               size = *count;
-       if (buf->sent + size != buf->sg.length || *count != size)
-               flags |= MSG_MORE;
-
-       res = iscsi_send(conn, buf, size, flags);
-       debug_tcp("sendpage: %d bytes, sent %d left %d sent %d res %d\n",
-                 size, buf->sent, *count, *sent, res);
-       if (res >= 0) {
-               *count -= res;
-               *sent += res;
-               if (size != res)
-                       return -EAGAIN;
-               return 0;
-       }
+       read_lock(&sk->sk_callback_lock);
 
-       return res;
-}
+       conn = (struct iscsi_conn*)sk->sk_user_data;
+       session = conn->session;
 
-static inline void
-iscsi_data_digest_init(struct iscsi_tcp_conn *tcp_conn,
-                     struct iscsi_tcp_cmd_task *tcp_ctask)
-{
-       crypto_hash_init(&tcp_conn->tx_hash);
-       tcp_ctask->digest_count = 4;
+       iscsi_sw_sk_state_check(sk);
+
+       tcp_conn = conn->dd_data;
+       tcp_sw_conn = tcp_conn->dd_data;
+       old_state_change = tcp_sw_conn->old_state_change;
+
+       read_unlock(&sk->sk_callback_lock);
+
+       old_state_change(sk);
 }
 
 /**
- * iscsi_solicit_data_cont - initialize next Data-Out
- * @conn: iscsi connection
- * @ctask: scsi command task
- * @r2t: R2T info
- * @left: bytes left to transfer
- *
- * Notes:
- *     Initialize next Data-Out within this R2T sequence and continue
- *     to process next Scatter-Gather element(if any) of this SCSI command.
- *
- *     Called under connection lock.
+ * iscsi_write_space - Called when more output buffer space is available
+ * @sk: socket space is available for
  **/
-static void
-iscsi_solicit_data_cont(struct iscsi_conn *conn, struct iscsi_cmd_task *ctask,
-                       struct iscsi_r2t_info *r2t, int left)
+static void iscsi_sw_tcp_write_space(struct sock *sk)
 {
-       struct iscsi_data *hdr;
-       int new_offset;
-
-       hdr = &r2t->dtask.hdr;
-       memset(hdr, 0, sizeof(struct iscsi_data));
-       hdr->ttt = r2t->ttt;
-       hdr->datasn = cpu_to_be32(r2t->solicit_datasn);
-       r2t->solicit_datasn++;
-       hdr->opcode = ISCSI_OP_SCSI_DATA_OUT;
-       memcpy(hdr->lun, ctask->hdr->lun, sizeof(hdr->lun));
-       hdr->itt = ctask->hdr->itt;
-       hdr->exp_statsn = r2t->exp_statsn;
-       new_offset = r2t->data_offset + r2t->sent;
-       hdr->offset = cpu_to_be32(new_offset);
-       if (left > conn->max_xmit_dlength) {
-               hton24(hdr->dlength, conn->max_xmit_dlength);
-               r2t->data_count = conn->max_xmit_dlength;
-       } else {
-               hton24(hdr->dlength, left);
-               r2t->data_count = left;
-               hdr->flags = ISCSI_FLAG_CMD_FINAL;
-       }
-       conn->dataout_pdus_cnt++;
-
-       iscsi_buf_init_iov(&r2t->headbuf, (char*)hdr,
-                          sizeof(struct iscsi_hdr));
-
-       if (iscsi_buf_left(&r2t->sendbuf))
-               return;
+       struct iscsi_conn *conn = (struct iscsi_conn*)sk->sk_user_data;
+       struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
+       struct iscsi_sw_tcp_conn *tcp_sw_conn = tcp_conn->dd_data;
 
-       iscsi_buf_init_sg(&r2t->sendbuf, r2t->sg);
-       r2t->sg += 1;
+       tcp_sw_conn->old_write_space(sk);
+       ISCSI_SW_TCP_DBG(conn, "iscsi_write_space\n");
+       iscsi_conn_queue_work(conn);
 }
 
-static void iscsi_set_padding(struct iscsi_tcp_cmd_task *tcp_ctask,
-                             unsigned long len)
+static void iscsi_sw_tcp_conn_set_callbacks(struct iscsi_conn *conn)
 {
-       tcp_ctask->pad_count = len & (ISCSI_PAD_LEN - 1);
-       if (!tcp_ctask->pad_count)
-               return;
+       struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
+       struct iscsi_sw_tcp_conn *tcp_sw_conn = tcp_conn->dd_data;
+       struct sock *sk = tcp_sw_conn->sock->sk;
 
-       tcp_ctask->pad_count = ISCSI_PAD_LEN - tcp_ctask->pad_count;
-       debug_scsi("write padding %d bytes\n", tcp_ctask->pad_count);
-       tcp_ctask->xmstate |= XMSTATE_W_PAD;
+       /* assign new callbacks */
+       write_lock_bh(&sk->sk_callback_lock);
+       sk->sk_user_data = conn;
+       tcp_sw_conn->old_data_ready = sk->sk_data_ready;
+       tcp_sw_conn->old_state_change = sk->sk_state_change;
+       tcp_sw_conn->old_write_space = sk->sk_write_space;
+       sk->sk_data_ready = iscsi_sw_tcp_data_ready;
+       sk->sk_state_change = iscsi_sw_tcp_state_change;
+       sk->sk_write_space = iscsi_sw_tcp_write_space;
+       write_unlock_bh(&sk->sk_callback_lock);
 }
 
-/**
- * iscsi_tcp_cmd_init - Initialize iSCSI SCSI_READ or SCSI_WRITE commands
- * @conn: iscsi connection
- * @ctask: scsi command task
- * @sc: scsi command
- **/
 static void
-iscsi_tcp_cmd_init(struct iscsi_cmd_task *ctask)
+iscsi_sw_tcp_conn_restore_callbacks(struct iscsi_conn *conn)
 {
-       struct iscsi_tcp_cmd_task *tcp_ctask = ctask->dd_data;
+       struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
+       struct iscsi_sw_tcp_conn *tcp_sw_conn = tcp_conn->dd_data;
+       struct sock *sk = tcp_sw_conn->sock->sk;
 
-       BUG_ON(__kfifo_len(tcp_ctask->r2tqueue));
-       tcp_ctask->xmstate = XMSTATE_CMD_HDR_INIT;
+       /* restore socket callbacks, see also: iscsi_conn_set_callbacks() */
+       write_lock_bh(&sk->sk_callback_lock);
+       sk->sk_user_data    = NULL;
+       sk->sk_data_ready   = tcp_sw_conn->old_data_ready;
+       sk->sk_state_change = tcp_sw_conn->old_state_change;
+       sk->sk_write_space  = tcp_sw_conn->old_write_space;
+       sk->sk_no_check  = 0;
+       write_unlock_bh(&sk->sk_callback_lock);
 }
 
 /**
- * iscsi_tcp_mtask_xmit - xmit management(immediate) task
- * @conn: iscsi connection
- * @mtask: task management task
+ * iscsi_sw_tcp_xmit_segment - transmit segment
+ * @tcp_conn: the iSCSI TCP connection
+ * @segment: the buffer to transmnit
  *
- * Notes:
- *     The function can return -EAGAIN in which case caller must
- *     call it again later, or recover. '0' return code means successful
- *     xmit.
+ * This function transmits as much of the buffer as
+ * the network layer will accept, and returns the number of
+ * bytes transmitted.
  *
- *     Management xmit state machine consists of these states:
- *             XMSTATE_IMM_HDR_INIT    - calculate digest of PDU Header
- *             XMSTATE_IMM_HDR         - PDU Header xmit in progress
- *             XMSTATE_IMM_DATA        - PDU Data xmit in progress
- *             XMSTATE_IDLE            - management PDU is done
- **/
-static int
-iscsi_tcp_mtask_xmit(struct iscsi_conn *conn, struct iscsi_mgmt_task *mtask)
-{
-       struct iscsi_tcp_mgmt_task *tcp_mtask = mtask->dd_data;
-       int rc;
-
-       debug_scsi("mtask deq [cid %d state %x itt 0x%x]\n",
-               conn->id, tcp_mtask->xmstate, mtask->itt);
-
-       if (tcp_mtask->xmstate & XMSTATE_IMM_HDR_INIT) {
-               iscsi_buf_init_iov(&tcp_mtask->headbuf, (char*)mtask->hdr,
-                                  sizeof(struct iscsi_hdr));
-
-               if (mtask->data_count) {
-                       tcp_mtask->xmstate |= XMSTATE_IMM_DATA;
-                       iscsi_buf_init_iov(&tcp_mtask->sendbuf,
-                                          (char*)mtask->data,
-                                          mtask->data_count);
+ * If CRC hashing is enabled, the function will compute the
+ * hash as it goes. When the entire segment has been transmitted,
+ * it will retrieve the hash value and send it as well.
+ */
+static int iscsi_sw_tcp_xmit_segment(struct iscsi_tcp_conn *tcp_conn,
+                                    struct iscsi_segment *segment)
+{
+       struct iscsi_sw_tcp_conn *tcp_sw_conn = tcp_conn->dd_data;
+       struct socket *sk = tcp_sw_conn->sock;
+       unsigned int copied = 0;
+       int r = 0;
+
+       while (!iscsi_tcp_segment_done(tcp_conn, segment, 0, r)) {
+               struct scatterlist *sg;
+               unsigned int offset, copy;
+               int flags = 0;
+
+               r = 0;
+               offset = segment->copied;
+               copy = segment->size - offset;
+
+               if (segment->total_copied + segment->size < segment->total_size)
+                       flags |= MSG_MORE;
+
+               /* Use sendpage if we can; else fall back to sendmsg */
+               if (!segment->data) {
+                       sg = segment->sg;
+                       offset += segment->sg_offset + sg->offset;
+                       r = tcp_sw_conn->sendpage(sk, sg_page(sg), offset,
+                                                 copy, flags);
+               } else {
+                       struct msghdr msg = { .msg_flags = flags };
+                       struct kvec iov = {
+                               .iov_base = segment->data + offset,
+                               .iov_len = copy
+                       };
+
+                       r = kernel_sendmsg(sk, &msg, &iov, 1, copy);
                }
 
-               if (conn->c_stage != ISCSI_CONN_INITIAL_STAGE &&
-                   conn->stop_stage != STOP_CONN_RECOVER &&
-                   conn->hdrdgst_en)
-                       iscsi_hdr_digest(conn, &tcp_mtask->headbuf,
-                                       (u8*)tcp_mtask->hdrext);
-
-               tcp_mtask->sent = 0;
-               tcp_mtask->xmstate &= ~XMSTATE_IMM_HDR_INIT;
-               tcp_mtask->xmstate |= XMSTATE_IMM_HDR;
-       }
-
-       if (tcp_mtask->xmstate & XMSTATE_IMM_HDR) {
-               rc = iscsi_sendhdr(conn, &tcp_mtask->headbuf,
-                                  mtask->data_count);
-               if (rc)
-                       return rc;
-               tcp_mtask->xmstate &= ~XMSTATE_IMM_HDR;
-       }
-
-       if (tcp_mtask->xmstate & XMSTATE_IMM_DATA) {
-               BUG_ON(!mtask->data_count);
-               tcp_mtask->xmstate &= ~XMSTATE_IMM_DATA;
-               /* FIXME: implement.
-                * Virtual buffer could be spreaded across multiple pages...
-                */
-               do {
-                       int rc;
-
-                       rc = iscsi_sendpage(conn, &tcp_mtask->sendbuf,
-                                       &mtask->data_count, &tcp_mtask->sent);
-                       if (rc) {
-                               tcp_mtask->xmstate |= XMSTATE_IMM_DATA;
-                               return rc;
-                       }
-               } while (mtask->data_count);
-       }
-
-       BUG_ON(tcp_mtask->xmstate != XMSTATE_IDLE);
-       if (mtask->hdr->itt == RESERVED_ITT) {
-               struct iscsi_session *session = conn->session;
-
-               spin_lock_bh(&session->lock);
-               list_del(&conn->mtask->running);
-               __kfifo_put(session->mgmtpool.queue, (void*)&conn->mtask,
-                           sizeof(void*));
-               spin_unlock_bh(&session->lock);
+               if (r < 0) {
+                       iscsi_tcp_segment_unmap(segment);
+                       return r;
+               }
+               copied += r;
        }
-       return 0;
+       return copied;
 }
 
-static int
-iscsi_send_cmd_hdr(struct iscsi_conn *conn, struct iscsi_cmd_task *ctask)
+/**
+ * iscsi_sw_tcp_xmit - TCP transmit
+ **/
+static int iscsi_sw_tcp_xmit(struct iscsi_conn *conn)
 {
-       struct scsi_cmnd *sc = ctask->sc;
-       struct iscsi_tcp_cmd_task *tcp_ctask = ctask->dd_data;
+       struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
+       struct iscsi_sw_tcp_conn *tcp_sw_conn = tcp_conn->dd_data;
+       struct iscsi_segment *segment = &tcp_sw_conn->out.segment;
+       unsigned int consumed = 0;
        int rc = 0;
 
-       if (tcp_ctask->xmstate & XMSTATE_CMD_HDR_INIT) {
-               tcp_ctask->sent = 0;
-               tcp_ctask->sg_count = 0;
-               tcp_ctask->exp_datasn = 0;
-
-               if (sc->sc_data_direction == DMA_TO_DEVICE) {
-                       struct scatterlist *sg = scsi_sglist(sc);
+       while (1) {
+               rc = iscsi_sw_tcp_xmit_segment(tcp_conn, segment);
+               /*
+                * We may not have been able to send data because the conn
+                * is getting stopped. libiscsi will know so propogate err
+                * for it to do the right thing.
+                */
+               if (rc == -EAGAIN)
+                       return rc;
+               else if (rc < 0) {
+                       rc = ISCSI_ERR_XMIT_FAILED;
+                       goto error;
+               } else if (rc == 0)
+                       break;
 
-                       iscsi_buf_init_sg(&tcp_ctask->sendbuf, sg);
-                       tcp_ctask->sg = sg + 1;
-                       tcp_ctask->bad_sg = sg + scsi_sg_count(sc);
+               consumed += rc;
 
-                       debug_scsi("cmd [itt 0x%x total %d imm_data %d "
-                                  "unsol count %d, unsol offset %d]\n",
-                                  ctask->itt, scsi_bufflen(sc),
-                                  ctask->imm_count, ctask->unsol_count,
-                                  ctask->unsol_offset);
+               if (segment->total_copied >= segment->total_size) {
+                       if (segment->done != NULL) {
+                               rc = segment->done(tcp_conn, segment);
+                               if (rc != 0)
+                                       goto error;
+                       }
                }
-
-               iscsi_buf_init_iov(&tcp_ctask->headbuf, (char*)ctask->hdr,
-                                 sizeof(struct iscsi_hdr));
-
-               if (conn->hdrdgst_en)
-                       iscsi_hdr_digest(conn, &tcp_ctask->headbuf,
-                                        (u8*)tcp_ctask->hdrext);
-               tcp_ctask->xmstate &= ~XMSTATE_CMD_HDR_INIT;
-               tcp_ctask->xmstate |= XMSTATE_CMD_HDR_XMIT;
        }
 
-       if (tcp_ctask->xmstate & XMSTATE_CMD_HDR_XMIT) {
-               rc = iscsi_sendhdr(conn, &tcp_ctask->headbuf, ctask->imm_count);
-               if (rc)
-                       return rc;
-               tcp_ctask->xmstate &= ~XMSTATE_CMD_HDR_XMIT;
-
-               if (sc->sc_data_direction != DMA_TO_DEVICE)
-                       return 0;
+       ISCSI_SW_TCP_DBG(conn, "xmit %d bytes\n", consumed);
 
-               if (ctask->imm_count) {
-                       tcp_ctask->xmstate |= XMSTATE_IMM_DATA;
-                       iscsi_set_padding(tcp_ctask, ctask->imm_count);
-
-                       if (ctask->conn->datadgst_en) {
-                               iscsi_data_digest_init(ctask->conn->dd_data,
-                                                      tcp_ctask);
-                               tcp_ctask->immdigest = 0;
-                       }
-               }
+       conn->txdata_octets += consumed;
+       return consumed;
 
-               if (ctask->unsol_count)
-                       tcp_ctask->xmstate |=
-                                       XMSTATE_UNS_HDR | XMSTATE_UNS_INIT;
-       }
-       return rc;
+error:
+       /* Transmit error. We could initiate error recovery
+        * here. */
+       ISCSI_SW_TCP_DBG(conn, "Error sending PDU, errno=%d\n", rc);
+       iscsi_conn_failure(conn, rc);
+       return -EIO;
 }
 
-static int
-iscsi_send_padding(struct iscsi_conn *conn, struct iscsi_cmd_task *ctask)
+/**
+ * iscsi_tcp_xmit_qlen - return the number of bytes queued for xmit
+ */
+static inline int iscsi_sw_tcp_xmit_qlen(struct iscsi_conn *conn)
 {
-       struct iscsi_tcp_cmd_task *tcp_ctask = ctask->dd_data;
        struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
-       int sent = 0, rc;
-
-       if (tcp_ctask->xmstate & XMSTATE_W_PAD) {
-               iscsi_buf_init_iov(&tcp_ctask->sendbuf, (char*)&tcp_ctask->pad,
-                                  tcp_ctask->pad_count);
-               if (conn->datadgst_en)
-                       crypto_hash_update(&tcp_conn->tx_hash,
-                                          &tcp_ctask->sendbuf.sg,
-                                          tcp_ctask->sendbuf.sg.length);
-       } else if (!(tcp_ctask->xmstate & XMSTATE_W_RESEND_PAD))
-               return 0;
+       struct iscsi_sw_tcp_conn *tcp_sw_conn = tcp_conn->dd_data;
+       struct iscsi_segment *segment = &tcp_sw_conn->out.segment;
 
-       tcp_ctask->xmstate &= ~XMSTATE_W_PAD;
-       tcp_ctask->xmstate &= ~XMSTATE_W_RESEND_PAD;
-       debug_scsi("sending %d pad bytes for itt 0x%x\n",
-                  tcp_ctask->pad_count, ctask->itt);
-       rc = iscsi_sendpage(conn, &tcp_ctask->sendbuf, &tcp_ctask->pad_count,
-                          &sent);
-       if (rc) {
-               debug_scsi("padding send failed %d\n", rc);
-               tcp_ctask->xmstate |= XMSTATE_W_RESEND_PAD;
-       }
-       return rc;
+       return segment->total_copied - segment->total_size;
 }
 
-static int
-iscsi_send_digest(struct iscsi_conn *conn, struct iscsi_cmd_task *ctask,
-                       struct iscsi_buf *buf, uint32_t *digest)
+static int iscsi_sw_tcp_pdu_xmit(struct iscsi_task *task)
 {
-       struct iscsi_tcp_cmd_task *tcp_ctask;
-       struct iscsi_tcp_conn *tcp_conn;
-       int rc, sent = 0;
-
-       if (!conn->datadgst_en)
-               return 0;
-
-       tcp_ctask = ctask->dd_data;
-       tcp_conn = conn->dd_data;
+       struct iscsi_conn *conn = task->conn;
+       int rc;
 
-       if (!(tcp_ctask->xmstate & XMSTATE_W_RESEND_DATA_DIGEST)) {
-               crypto_hash_final(&tcp_conn->tx_hash, (u8*)digest);
-               iscsi_buf_init_iov(buf, (char*)digest, 4);
+       while (iscsi_sw_tcp_xmit_qlen(conn)) {
+               rc = iscsi_sw_tcp_xmit(conn);
+               if (rc == 0)
+                       return -EAGAIN;
+               if (rc < 0)
+                       return rc;
        }
-       tcp_ctask->xmstate &= ~XMSTATE_W_RESEND_DATA_DIGEST;
 
-       rc = iscsi_sendpage(conn, buf, &tcp_ctask->digest_count, &sent);
-       if (!rc)
-               debug_scsi("sent digest 0x%x for itt 0x%x\n", *digest,
-                         ctask->itt);
-       else {
-               debug_scsi("sending digest 0x%x failed for itt 0x%x!\n",
-                         *digest, ctask->itt);
-               tcp_ctask->xmstate |= XMSTATE_W_RESEND_DATA_DIGEST;
-       }
-       return rc;
+       return 0;
 }
 
-static int
-iscsi_send_data(struct iscsi_cmd_task *ctask, struct iscsi_buf *sendbuf,
-               struct scatterlist **sg, int *sent, int *count,
-               struct iscsi_buf *digestbuf, uint32_t *digest)
+/*
+ * This is called when we're done sending the header.
+ * Simply copy the data_segment to the send segment, and return.
+ */
+static int iscsi_sw_tcp_send_hdr_done(struct iscsi_tcp_conn *tcp_conn,
+                                     struct iscsi_segment *segment)
 {
-       struct iscsi_tcp_cmd_task *tcp_ctask = ctask->dd_data;
-       struct iscsi_conn *conn = ctask->conn;
-       struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
-       int rc, buf_sent, offset;
-
-       while (*count) {
-               buf_sent = 0;
-               offset = sendbuf->sent;
-
-               rc = iscsi_sendpage(conn, sendbuf, count, &buf_sent);
-               *sent = *sent + buf_sent;
-               if (buf_sent && conn->datadgst_en)
-                       partial_sg_digest_update(&tcp_conn->tx_hash,
-                               &sendbuf->sg, sendbuf->sg.offset + offset,
-                               buf_sent);
-               if (!iscsi_buf_left(sendbuf) && *sg != tcp_ctask->bad_sg) {
-                       iscsi_buf_init_sg(sendbuf, *sg);
-                       *sg = *sg + 1;
-               }
-
-               if (rc)
-                       return rc;
-       }
+       struct iscsi_sw_tcp_conn *tcp_sw_conn = tcp_conn->dd_data;
 
-       rc = iscsi_send_padding(conn, ctask);
-       if (rc)
-               return rc;
-
-       return iscsi_send_digest(conn, ctask, digestbuf, digest);
+       tcp_sw_conn->out.segment = tcp_sw_conn->out.data_segment;
+       ISCSI_SW_TCP_DBG(tcp_conn->iscsi_conn,
+                        "Header done. Next segment size %u total_size %u\n",
+                        tcp_sw_conn->out.segment.size,
+                        tcp_sw_conn->out.segment.total_size);
+       return 0;
 }
 
-static int
-iscsi_send_unsol_hdr(struct iscsi_conn *conn, struct iscsi_cmd_task *ctask)
+static void iscsi_sw_tcp_send_hdr_prep(struct iscsi_conn *conn, void *hdr,
+                                      size_t hdrlen)
 {
-       struct iscsi_tcp_cmd_task *tcp_ctask = ctask->dd_data;
-       struct iscsi_data_task *dtask;
-       int rc;
-
-       tcp_ctask->xmstate |= XMSTATE_UNS_DATA;
-       if (tcp_ctask->xmstate & XMSTATE_UNS_INIT) {
-               dtask = &tcp_ctask->unsol_dtask;
+       struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
+       struct iscsi_sw_tcp_conn *tcp_sw_conn = tcp_conn->dd_data;
 
-               iscsi_prep_unsolicit_data_pdu(ctask, &dtask->hdr);
-               iscsi_buf_init_iov(&tcp_ctask->headbuf, (char*)&dtask->hdr,
-                                  sizeof(struct iscsi_hdr));
-               if (conn->hdrdgst_en)
-                       iscsi_hdr_digest(conn, &tcp_ctask->headbuf,
-                                       (u8*)dtask->hdrext);
+       ISCSI_SW_TCP_DBG(conn, "%s\n", conn->hdrdgst_en ?
+                        "digest enabled" : "digest disabled");
 
-               tcp_ctask->xmstate &= ~XMSTATE_UNS_INIT;
-               iscsi_set_padding(tcp_ctask, ctask->data_count);
-       }
+       /* Clear the data segment - needs to be filled in by the
+        * caller using iscsi_tcp_send_data_prep() */
+       memset(&tcp_sw_conn->out.data_segment, 0,
+              sizeof(struct iscsi_segment));
 
-       rc = iscsi_sendhdr(conn, &tcp_ctask->headbuf, ctask->data_count);
-       if (rc) {
-               tcp_ctask->xmstate &= ~XMSTATE_UNS_DATA;
-               tcp_ctask->xmstate |= XMSTATE_UNS_HDR;
-               return rc;
+       /* If header digest is enabled, compute the CRC and
+        * place the digest into the same buffer. We make
+        * sure that both iscsi_tcp_task and mtask have
+        * sufficient room.
+        */
+       if (conn->hdrdgst_en) {
+               iscsi_tcp_dgst_header(&tcp_sw_conn->tx_hash, hdr, hdrlen,
+                                     hdr + hdrlen);
+               hdrlen += ISCSI_DIGEST_SIZE;
        }
 
-       if (conn->datadgst_en) {
-               dtask = &tcp_ctask->unsol_dtask;
-               iscsi_data_digest_init(ctask->conn->dd_data, tcp_ctask);
-               dtask->digest = 0;
-       }
+       /* Remember header pointer for later, when we need
+        * to decide whether there's a payload to go along
+        * with the header. */
+       tcp_sw_conn->out.hdr = hdr;
 
-       debug_scsi("uns dout [itt 0x%x dlen %d sent %d]\n",
-                  ctask->itt, ctask->unsol_count, tcp_ctask->sent);
-       return 0;
+       iscsi_segment_init_linear(&tcp_sw_conn->out.segment, hdr, hdrlen,
+                                 iscsi_sw_tcp_send_hdr_done, NULL);
 }
 
+/*
+ * Prepare the send buffer for the payload data.
+ * Padding and checksumming will all be taken care
+ * of by the iscsi_segment routines.
+ */
 static int
-iscsi_send_unsol_pdu(struct iscsi_conn *conn, struct iscsi_cmd_task *ctask)
+iscsi_sw_tcp_send_data_prep(struct iscsi_conn *conn, struct scatterlist *sg,
+                           unsigned int count, unsigned int offset,
+                           unsigned int len)
 {
-       struct iscsi_tcp_cmd_task *tcp_ctask = ctask->dd_data;
-       int rc;
+       struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
+       struct iscsi_sw_tcp_conn *tcp_sw_conn = tcp_conn->dd_data;
+       struct hash_desc *tx_hash = NULL;
+       unsigned int hdr_spec_len;
 
-       if (tcp_ctask->xmstate & XMSTATE_UNS_HDR) {
-               BUG_ON(!ctask->unsol_count);
-               tcp_ctask->xmstate &= ~XMSTATE_UNS_HDR;
-send_hdr:
-               rc = iscsi_send_unsol_hdr(conn, ctask);
-               if (rc)
-                       return rc;
-       }
+       ISCSI_SW_TCP_DBG(conn, "offset=%d, datalen=%d %s\n", offset, len,
+                        conn->datadgst_en ?
+                        "digest enabled" : "digest disabled");
 
-       if (tcp_ctask->xmstate & XMSTATE_UNS_DATA) {
-               struct iscsi_data_task *dtask = &tcp_ctask->unsol_dtask;
-               int start = tcp_ctask->sent;
+       /* Make sure the datalen matches what the caller
+          said he would send. */
+       hdr_spec_len = ntoh24(tcp_sw_conn->out.hdr->dlength);
+       WARN_ON(iscsi_padded(len) != iscsi_padded(hdr_spec_len));
 
-               rc = iscsi_send_data(ctask, &tcp_ctask->sendbuf, &tcp_ctask->sg,
-                                    &tcp_ctask->sent, &ctask->data_count,
-                                    &dtask->digestbuf, &dtask->digest);
-               ctask->unsol_count -= tcp_ctask->sent - start;
-               if (rc)
-                       return rc;
-               tcp_ctask->xmstate &= ~XMSTATE_UNS_DATA;
-               /*
-                * Done with the Data-Out. Next, check if we need
-                * to send another unsolicited Data-Out.
-                */
-               if (ctask->unsol_count) {
-                       debug_scsi("sending more uns\n");
-                       tcp_ctask->xmstate |= XMSTATE_UNS_INIT;
-                       goto send_hdr;
-               }
-       }
-       return 0;
+       if (conn->datadgst_en)
+               tx_hash = &tcp_sw_conn->tx_hash;
+
+       return iscsi_segment_seek_sg(&tcp_sw_conn->out.data_segment,
+                                    sg, count, offset, len,
+                                    NULL, tx_hash);
 }
 
-static int iscsi_send_sol_pdu(struct iscsi_conn *conn,
-                             struct iscsi_cmd_task *ctask)
+static void
+iscsi_sw_tcp_send_linear_data_prep(struct iscsi_conn *conn, void *data,
+                                  size_t len)
 {
-       struct iscsi_tcp_cmd_task *tcp_ctask = ctask->dd_data;
-       struct iscsi_session *session = conn->session;
-       struct iscsi_r2t_info *r2t;
-       struct iscsi_data_task *dtask;
-       int left, rc;
-
-       if (tcp_ctask->xmstate & XMSTATE_SOL_HDR_INIT) {
-               if (!tcp_ctask->r2t) {
-                       spin_lock_bh(&session->lock);
-                       __kfifo_get(tcp_ctask->r2tqueue, (void*)&tcp_ctask->r2t,
-                                   sizeof(void*));
-                       spin_unlock_bh(&session->lock);
-               }
-send_hdr:
-               r2t = tcp_ctask->r2t;
-               dtask = &r2t->dtask;
-
-               if (conn->hdrdgst_en)
-                       iscsi_hdr_digest(conn, &r2t->headbuf,
-                                       (u8*)dtask->hdrext);
-               tcp_ctask->xmstate &= ~XMSTATE_SOL_HDR_INIT;
-               tcp_ctask->xmstate |= XMSTATE_SOL_HDR;
-       }
-
-       if (tcp_ctask->xmstate & XMSTATE_SOL_HDR) {
-               r2t = tcp_ctask->r2t;
-               dtask = &r2t->dtask;
-
-               rc = iscsi_sendhdr(conn, &r2t->headbuf, r2t->data_count);
-               if (rc)
-                       return rc;
-               tcp_ctask->xmstate &= ~XMSTATE_SOL_HDR;
-               tcp_ctask->xmstate |= XMSTATE_SOL_DATA;
-
-               if (conn->datadgst_en) {
-                       iscsi_data_digest_init(conn->dd_data, tcp_ctask);
-                       dtask->digest = 0;
-               }
-
-               iscsi_set_padding(tcp_ctask, r2t->data_count);
-               debug_scsi("sol dout [dsn %d itt 0x%x dlen %d sent %d]\n",
-                       r2t->solicit_datasn - 1, ctask->itt, r2t->data_count,
-                       r2t->sent);
-       }
+       struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
+       struct iscsi_sw_tcp_conn *tcp_sw_conn = tcp_conn->dd_data;
+       struct hash_desc *tx_hash = NULL;
+       unsigned int hdr_spec_len;
 
-       if (tcp_ctask->xmstate & XMSTATE_SOL_DATA) {
-               r2t = tcp_ctask->r2t;
-               dtask = &r2t->dtask;
+       ISCSI_SW_TCP_DBG(conn, "datalen=%zd %s\n", len, conn->datadgst_en ?
+                        "digest enabled" : "digest disabled");
 
-               rc = iscsi_send_data(ctask, &r2t->sendbuf, &r2t->sg,
-                                    &r2t->sent, &r2t->data_count,
-                                    &dtask->digestbuf, &dtask->digest);
-               if (rc)
-                       return rc;
-               tcp_ctask->xmstate &= ~XMSTATE_SOL_DATA;
+       /* Make sure the datalen matches what the caller
+          said he would send. */
+       hdr_spec_len = ntoh24(tcp_sw_conn->out.hdr->dlength);
+       WARN_ON(iscsi_padded(len) != iscsi_padded(hdr_spec_len));
 
-               /*
-                * Done with this Data-Out. Next, check if we have
-                * to send another Data-Out for this R2T.
-                */
-               BUG_ON(r2t->data_length - r2t->sent < 0);
-               left = r2t->data_length - r2t->sent;
-               if (left) {
-                       iscsi_solicit_data_cont(conn, ctask, r2t, left);
-                       goto send_hdr;
-               }
+       if (conn->datadgst_en)
+               tx_hash = &tcp_sw_conn->tx_hash;
 
-               /*
-                * Done with this R2T. Check if there are more
-                * outstanding R2Ts ready to be processed.
-                */
-               spin_lock_bh(&session->lock);
-               tcp_ctask->r2t = NULL;
-               __kfifo_put(tcp_ctask->r2tpool.queue, (void*)&r2t,
-                           sizeof(void*));
-               if (__kfifo_get(tcp_ctask->r2tqueue, (void*)&r2t,
-                               sizeof(void*))) {
-                       tcp_ctask->r2t = r2t;
-                       spin_unlock_bh(&session->lock);
-                       goto send_hdr;
-               }
-               spin_unlock_bh(&session->lock);
-       }
-       return 0;
+       iscsi_segment_init_linear(&tcp_sw_conn->out.data_segment,
+                               data, len, NULL, tx_hash);
 }
 
-/**
- * iscsi_tcp_ctask_xmit - xmit normal PDU task
- * @conn: iscsi connection
- * @ctask: iscsi command task
- *
- * Notes:
- *     The function can return -EAGAIN in which case caller must
- *     call it again later, or recover. '0' return code means successful
- *     xmit.
- *     The function is devided to logical helpers (above) for the different
- *     xmit stages.
- *
- *iscsi_send_cmd_hdr()
- *     XMSTATE_CMD_HDR_INIT - prepare Header and Data buffers Calculate
- *                            Header Digest
- *     XMSTATE_CMD_HDR_XMIT - Transmit header in progress
- *
- *iscsi_send_padding
- *     XMSTATE_W_PAD        - Prepare and send pading
- *     XMSTATE_W_RESEND_PAD - retry send pading
- *
- *iscsi_send_digest
- *     XMSTATE_W_RESEND_DATA_DIGEST - Finalize and send Data Digest
- *     XMSTATE_W_RESEND_DATA_DIGEST - retry sending digest
- *
- *iscsi_send_unsol_hdr
- *     XMSTATE_UNS_INIT     - prepare un-solicit data header and digest
- *     XMSTATE_UNS_HDR      - send un-solicit header
- *
- *iscsi_send_unsol_pdu
- *     XMSTATE_UNS_DATA     - send un-solicit data in progress
- *
- *iscsi_send_sol_pdu
- *     XMSTATE_SOL_HDR_INIT - solicit data header and digest initialize
- *     XMSTATE_SOL_HDR      - send solicit header
- *     XMSTATE_SOL_DATA     - send solicit data
- *
- *iscsi_tcp_ctask_xmit
- *     XMSTATE_IMM_DATA     - xmit managment data (??)
- **/
-static int
-iscsi_tcp_ctask_xmit(struct iscsi_conn *conn, struct iscsi_cmd_task *ctask)
+static int iscsi_sw_tcp_pdu_init(struct iscsi_task *task,
+                                unsigned int offset, unsigned int count)
 {
-       struct iscsi_tcp_cmd_task *tcp_ctask = ctask->dd_data;
-       int rc = 0;
+       struct iscsi_conn *conn = task->conn;
+       int err = 0;
 
-       debug_scsi("ctask deq [cid %d xmstate %x itt 0x%x]\n",
-               conn->id, tcp_ctask->xmstate, ctask->itt);
+       iscsi_sw_tcp_send_hdr_prep(conn, task->hdr, task->hdr_len);
 
-       rc = iscsi_send_cmd_hdr(conn, ctask);
-       if (rc)
-               return rc;
-       if (ctask->sc->sc_data_direction != DMA_TO_DEVICE)
+       if (!count)
                return 0;
 
-       if (tcp_ctask->xmstate & XMSTATE_IMM_DATA) {
-               rc = iscsi_send_data(ctask, &tcp_ctask->sendbuf, &tcp_ctask->sg,
-                                    &tcp_ctask->sent, &ctask->imm_count,
-                                    &tcp_ctask->immbuf, &tcp_ctask->immdigest);
-               if (rc)
-                       return rc;
-               tcp_ctask->xmstate &= ~XMSTATE_IMM_DATA;
+       if (!task->sc)
+               iscsi_sw_tcp_send_linear_data_prep(conn, task->data, count);
+       else {
+               struct scsi_data_buffer *sdb = scsi_out(task->sc);
+
+               err = iscsi_sw_tcp_send_data_prep(conn, sdb->table.sgl,
+                                                 sdb->table.nents, offset,
+                                                 count);
        }
 
-       rc = iscsi_send_unsol_pdu(conn, ctask);
-       if (rc)
-               return rc;
+       if (err) {
+               /* got invalid offset/len */
+               return -EIO;
+       }
+       return 0;
+}
 
-       rc = iscsi_send_sol_pdu(conn, ctask);
-       if (rc)
-               return rc;
+static int iscsi_sw_tcp_pdu_alloc(struct iscsi_task *task, uint8_t opcode)
+{
+       struct iscsi_tcp_task *tcp_task = task->dd_data;
 
-       return rc;
+       task->hdr = task->dd_data + sizeof(*tcp_task);
+       task->hdr_max = sizeof(struct iscsi_sw_tcp_hdrbuf) - ISCSI_DIGEST_SIZE;
+       return 0;
 }
 
 static struct iscsi_cls_conn *
-iscsi_tcp_conn_create(struct iscsi_cls_session *cls_session, uint32_t conn_idx)
+iscsi_sw_tcp_conn_create(struct iscsi_cls_session *cls_session,
+                        uint32_t conn_idx)
 {
        struct iscsi_conn *conn;
        struct iscsi_cls_conn *cls_conn;
        struct iscsi_tcp_conn *tcp_conn;
+       struct iscsi_sw_tcp_conn *tcp_sw_conn;
 
-       cls_conn = iscsi_conn_setup(cls_session, conn_idx);
+       cls_conn = iscsi_tcp_conn_setup(cls_session, sizeof(*tcp_sw_conn),
+                                       conn_idx);
        if (!cls_conn)
                return NULL;
        conn = cls_conn->dd_data;
-       /*
-        * due to strange issues with iser these are not set
-        * in iscsi_conn_setup
-        */
-       conn->max_recv_dlength = ISCSI_DEF_MAX_RECV_SEG_LEN;
-
-       tcp_conn = kzalloc(sizeof(*tcp_conn), GFP_KERNEL);
-       if (!tcp_conn)
-               goto tcp_conn_alloc_fail;
-
-       conn->dd_data = tcp_conn;
-       tcp_conn->iscsi_conn = conn;
-
-       tcp_conn->tx_hash.tfm = crypto_alloc_hash("crc32c", 0,
-                                                 CRYPTO_ALG_ASYNC);
-       tcp_conn->tx_hash.flags = 0;
-       if (IS_ERR(tcp_conn->tx_hash.tfm)) {
-               printk(KERN_ERR "Could not create connection due to crc32c "
-                      "loading error %ld. Make sure the crc32c module is "
-                      "built as a module or into the kernel\n",
-                       PTR_ERR(tcp_conn->tx_hash.tfm));
-               goto free_tcp_conn;
-       }
-
-       tcp_conn->rx_hash.tfm = crypto_alloc_hash("crc32c", 0,
-                                                 CRYPTO_ALG_ASYNC);
-       tcp_conn->rx_hash.flags = 0;
-       if (IS_ERR(tcp_conn->rx_hash.tfm)) {
-               printk(KERN_ERR "Could not create connection due to crc32c "
-                      "loading error %ld. Make sure the crc32c module is "
-                      "built as a module or into the kernel\n",
-                       PTR_ERR(tcp_conn->rx_hash.tfm));
+       tcp_conn = conn->dd_data;
+       tcp_sw_conn = tcp_conn->dd_data;
+
+       tcp_sw_conn->tx_hash.tfm = crypto_alloc_hash("crc32c", 0,
+                                                    CRYPTO_ALG_ASYNC);
+       tcp_sw_conn->tx_hash.flags = 0;
+       if (IS_ERR(tcp_sw_conn->tx_hash.tfm))
+               goto free_conn;
+
+       tcp_sw_conn->rx_hash.tfm = crypto_alloc_hash("crc32c", 0,
+                                                    CRYPTO_ALG_ASYNC);
+       tcp_sw_conn->rx_hash.flags = 0;
+       if (IS_ERR(tcp_sw_conn->rx_hash.tfm))
                goto free_tx_tfm;
-       }
+       tcp_conn->rx_hash = &tcp_sw_conn->rx_hash;
 
        return cls_conn;
 
 free_tx_tfm:
-       crypto_free_hash(tcp_conn->tx_hash.tfm);
-free_tcp_conn:
-       kfree(tcp_conn);
-tcp_conn_alloc_fail:
-       iscsi_conn_teardown(cls_conn);
+       crypto_free_hash(tcp_sw_conn->tx_hash.tfm);
+free_conn:
+       iscsi_conn_printk(KERN_ERR, conn,
+                         "Could not create connection due to crc32c "
+                         "loading error. Make sure the crc32c "
+                         "module is built as a module or into the "
+                         "kernel\n");
+       iscsi_tcp_conn_teardown(cls_conn);
        return NULL;
 }
 
-static void
-iscsi_tcp_release_conn(struct iscsi_conn *conn)
+static void iscsi_sw_tcp_release_conn(struct iscsi_conn *conn)
 {
        struct iscsi_session *session = conn->session;
        struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
-       struct socket *sock = tcp_conn->sock;
+       struct iscsi_sw_tcp_conn *tcp_sw_conn = tcp_conn->dd_data;
+       struct socket *sock = tcp_sw_conn->sock;
 
        if (!sock)
                return;
 
        sock_hold(sock->sk);
-       iscsi_conn_restore_callbacks(tcp_conn);
+       iscsi_sw_tcp_conn_restore_callbacks(conn);
        sock_put(sock->sk);
 
        spin_lock_bh(&session->lock);
-       tcp_conn->sock = NULL;
-       conn->recv_lock = NULL;
+       tcp_sw_conn->sock = NULL;
        spin_unlock_bh(&session->lock);
        sockfd_put(sock);
 }
 
-static void
-iscsi_tcp_conn_destroy(struct iscsi_cls_conn *cls_conn)
+static void iscsi_sw_tcp_conn_destroy(struct iscsi_cls_conn *cls_conn)
 {
        struct iscsi_conn *conn = cls_conn->dd_data;
        struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
+       struct iscsi_sw_tcp_conn *tcp_sw_conn = tcp_conn->dd_data;
 
-       iscsi_tcp_release_conn(conn);
-       iscsi_conn_teardown(cls_conn);
+       iscsi_sw_tcp_release_conn(conn);
 
-       if (tcp_conn->tx_hash.tfm)
-               crypto_free_hash(tcp_conn->tx_hash.tfm);
-       if (tcp_conn->rx_hash.tfm)
-               crypto_free_hash(tcp_conn->rx_hash.tfm);
+       if (tcp_sw_conn->tx_hash.tfm)
+               crypto_free_hash(tcp_sw_conn->tx_hash.tfm);
+       if (tcp_sw_conn->rx_hash.tfm)
+               crypto_free_hash(tcp_sw_conn->rx_hash.tfm);
 
-       kfree(tcp_conn);
+       iscsi_tcp_conn_teardown(cls_conn);
 }
 
-static void
-iscsi_tcp_conn_stop(struct iscsi_cls_conn *cls_conn, int flag)
+static void iscsi_sw_tcp_conn_stop(struct iscsi_cls_conn *cls_conn, int flag)
 {
        struct iscsi_conn *conn = cls_conn->dd_data;
+       struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
+       struct iscsi_sw_tcp_conn *tcp_sw_conn = tcp_conn->dd_data;
+       struct socket *sock = tcp_sw_conn->sock;
+
+       /* userspace may have goofed up and not bound us */
+       if (!sock)
+               return;
+       /*
+        * Make sure our recv side is stopped.
+        * Older tools called conn stop before ep_disconnect
+        * so IO could still be coming in.
+        */
+       write_lock_bh(&tcp_sw_conn->sock->sk->sk_callback_lock);
+       set_bit(ISCSI_SUSPEND_BIT, &conn->suspend_rx);
+       write_unlock_bh(&tcp_sw_conn->sock->sk->sk_callback_lock);
+
+       if (sk_sleep(sock->sk) && waitqueue_active(sk_sleep(sock->sk))) {
+               sock->sk->sk_err = EIO;
+               wake_up_interruptible(sk_sleep(sock->sk));
+       }
 
        iscsi_conn_stop(cls_conn, flag);
-       iscsi_tcp_release_conn(conn);
+       iscsi_sw_tcp_release_conn(conn);
 }
 
-static int iscsi_tcp_get_addr(struct iscsi_conn *conn, struct socket *sock,
-                             char *buf, int *port,
-                             int (*getname)(struct socket *, struct sockaddr *,
-                                       int *addrlen))
+static int iscsi_sw_tcp_get_addr(struct iscsi_conn *conn, struct socket *sock,
+                                char *buf, int *port,
+                                int (*getname)(struct socket *,
+                                               struct sockaddr *,
+                                               int *addrlen))
 {
        struct sockaddr_storage *addr;
        struct sockaddr_in6 *sin6;
@@ -1889,14 +634,14 @@ static int iscsi_tcp_get_addr(struct iscsi_conn *conn, struct socket *sock,
        case AF_INET:
                sin = (struct sockaddr_in *)addr;
                spin_lock_bh(&conn->session->lock);
-               sprintf(buf, NIPQUAD_FMT, NIPQUAD(sin->sin_addr.s_addr));
+               sprintf(buf, "%pI4", &sin->sin_addr.s_addr);
                *port = be16_to_cpu(sin->sin_port);
                spin_unlock_bh(&conn->session->lock);
                break;
        case AF_INET6:
                sin6 = (struct sockaddr_in6 *)addr;
                spin_lock_bh(&conn->session->lock);
-               sprintf(buf, NIP6_FMT, NIP6(sin6->sin6_addr));
+               sprintf(buf, "%pI6", &sin6->sin6_addr);
                *port = be16_to_cpu(sin6->sin6_port);
                spin_unlock_bh(&conn->session->lock);
                break;
@@ -1907,12 +652,15 @@ free_addr:
 }
 
 static int
-iscsi_tcp_conn_bind(struct iscsi_cls_session *cls_session,
-                   struct iscsi_cls_conn *cls_conn, uint64_t transport_eph,
-                   int is_leading)
+iscsi_sw_tcp_conn_bind(struct iscsi_cls_session *cls_session,
+                      struct iscsi_cls_conn *cls_conn, uint64_t transport_eph,
+                      int is_leading)
 {
+       struct Scsi_Host *shost = iscsi_session_to_shost(cls_session);
+       struct iscsi_host *ihost = shost_priv(shost);
        struct iscsi_conn *conn = cls_conn->dd_data;
        struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
+       struct iscsi_sw_tcp_conn *tcp_sw_conn = tcp_conn->dd_data;
        struct sock *sk;
        struct socket *sock;
        int err;
@@ -1920,7 +668,8 @@ iscsi_tcp_conn_bind(struct iscsi_cls_session *cls_session,
        /* lookup for existing socket */
        sock = sockfd_lookup((int)transport_eph, &err);
        if (!sock) {
-               printk(KERN_ERR "iscsi_tcp: sockfd_lookup failed %d\n", err);
+               iscsi_conn_printk(KERN_ERR, conn,
+                                 "sockfd_lookup failed %d\n", err);
                return -EEXIST;
        }
        /*
@@ -1928,13 +677,13 @@ iscsi_tcp_conn_bind(struct iscsi_cls_session *cls_session,
         * userspace may still want to query the values since we will
         * be using them for the reconnect
         */
-       err = iscsi_tcp_get_addr(conn, sock, conn->portal_address,
-                                &conn->portal_port, kernel_getpeername);
+       err = iscsi_sw_tcp_get_addr(conn, sock, conn->portal_address,
+                                   &conn->portal_port, kernel_getpeername);
        if (err)
                goto free_socket;
 
-       err = iscsi_tcp_get_addr(conn, sock, conn->local_address,
-                               &conn->local_port, kernel_getsockname);
+       err = iscsi_sw_tcp_get_addr(conn, sock, ihost->local_address,
+                                   &ihost->local_port, kernel_getsockname);
        if (err)
                goto free_socket;
 
@@ -1943,7 +692,7 @@ iscsi_tcp_conn_bind(struct iscsi_cls_session *cls_session,
                goto free_socket;
 
        /* bind iSCSI connection and socket */
-       tcp_conn->sock = sock;
+       tcp_sw_conn->sock = sock;
 
        /* setup Socket parameters */
        sk = sock->sk;
@@ -1951,15 +700,8 @@ iscsi_tcp_conn_bind(struct iscsi_cls_session *cls_session,
        sk->sk_sndtimeo = 15 * HZ; /* FIXME: make it configurable */
        sk->sk_allocation = GFP_ATOMIC;
 
-       /* FIXME: disable Nagle's algorithm */
-
-       /*
-        * Intercept TCP callbacks for sendfile like receive
-        * processing.
-        */
-       conn->recv_lock = &sk->sk_callback_lock;
-       iscsi_conn_set_callbacks(conn);
-       tcp_conn->sendpage = tcp_conn->sock->ops->sendpage;
+       iscsi_sw_tcp_conn_set_callbacks(conn);
+       tcp_sw_conn->sendpage = tcp_sw_conn->sock->ops->sendpage;
        /*
         * set receive state machine into initial state
         */
@@ -1971,86 +713,14 @@ free_socket:
        return err;
 }
 
-/* called with host lock */
-static void
-iscsi_tcp_mgmt_init(struct iscsi_conn *conn, struct iscsi_mgmt_task *mtask)
-{
-       struct iscsi_tcp_mgmt_task *tcp_mtask = mtask->dd_data;
-       tcp_mtask->xmstate = XMSTATE_IMM_HDR_INIT;
-}
-
-static int
-iscsi_r2tpool_alloc(struct iscsi_session *session)
-{
-       int i;
-       int cmd_i;
-
-       /*
-        * initialize per-task: R2T pool and xmit queue
-        */
-       for (cmd_i = 0; cmd_i < session->cmds_max; cmd_i++) {
-               struct iscsi_cmd_task *ctask = session->cmds[cmd_i];
-               struct iscsi_tcp_cmd_task *tcp_ctask = ctask->dd_data;
-
-               /*
-                * pre-allocated x4 as much r2ts to handle race when
-                * target acks DataOut faster than we data_xmit() queues
-                * could replenish r2tqueue.
-                */
-
-               /* R2T pool */
-               if (iscsi_pool_init(&tcp_ctask->r2tpool, session->max_r2t * 4,
-                                   (void***)&tcp_ctask->r2ts,
-                                   sizeof(struct iscsi_r2t_info))) {
-                       goto r2t_alloc_fail;
-               }
-
-               /* R2T xmit queue */
-               tcp_ctask->r2tqueue = kfifo_alloc(
-                     session->max_r2t * 4 * sizeof(void*), GFP_KERNEL, NULL);
-               if (tcp_ctask->r2tqueue == ERR_PTR(-ENOMEM)) {
-                       iscsi_pool_free(&tcp_ctask->r2tpool,
-                                       (void**)tcp_ctask->r2ts);
-                       goto r2t_alloc_fail;
-               }
-       }
-
-       return 0;
-
-r2t_alloc_fail:
-       for (i = 0; i < cmd_i; i++) {
-               struct iscsi_cmd_task *ctask = session->cmds[i];
-               struct iscsi_tcp_cmd_task *tcp_ctask = ctask->dd_data;
-
-               kfifo_free(tcp_ctask->r2tqueue);
-               iscsi_pool_free(&tcp_ctask->r2tpool,
-                               (void**)tcp_ctask->r2ts);
-       }
-       return -ENOMEM;
-}
-
-static void
-iscsi_r2tpool_free(struct iscsi_session *session)
-{
-       int i;
-
-       for (i = 0; i < session->cmds_max; i++) {
-               struct iscsi_cmd_task *ctask = session->cmds[i];
-               struct iscsi_tcp_cmd_task *tcp_ctask = ctask->dd_data;
-
-               kfifo_free(tcp_ctask->r2tqueue);
-               iscsi_pool_free(&tcp_ctask->r2tpool,
-                               (void**)tcp_ctask->r2ts);
-       }
-}
-
-static int
-iscsi_conn_set_param(struct iscsi_cls_conn *cls_conn, enum iscsi_param param,
-                    char *buf, int buflen)
+static int iscsi_sw_tcp_conn_set_param(struct iscsi_cls_conn *cls_conn,
+                                      enum iscsi_param param, char *buf,
+                                      int buflen)
 {
        struct iscsi_conn *conn = cls_conn->dd_data;
        struct iscsi_session *session = conn->session;
        struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
+       struct iscsi_sw_tcp_conn *tcp_sw_conn = tcp_conn->dd_data;
        int value;
 
        switch(param) {
@@ -2059,18 +729,18 @@ iscsi_conn_set_param(struct iscsi_cls_conn *cls_conn, enum iscsi_param param,
                break;
        case ISCSI_PARAM_DATADGST_EN:
                iscsi_set_param(cls_conn, param, buf, buflen);
-               tcp_conn->sendpage = conn->datadgst_en ?
-                       sock_no_sendpage : tcp_conn->sock->ops->sendpage;
+               tcp_sw_conn->sendpage = conn->datadgst_en ?
+                       sock_no_sendpage : tcp_sw_conn->sock->ops->sendpage;
                break;
        case ISCSI_PARAM_MAX_R2T:
                sscanf(buf, "%d", &value);
-               if (session->max_r2t == roundup_pow_of_two(value))
+               if (value <= 0 || !is_power_of_2(value))
+                       return -EINVAL;
+               if (session->max_r2t == value)
                        break;
-               iscsi_r2tpool_free(session);
+               iscsi_tcp_r2tpool_free(session);
                iscsi_set_param(cls_conn, param, buf, buflen);
-               if (session->max_r2t & (session->max_r2t - 1))
-                       session->max_r2t = roundup_pow_of_two(session->max_r2t);
-               if (iscsi_r2tpool_alloc(session))
+               if (iscsi_tcp_r2tpool_alloc(session))
                        return -ENOMEM;
                break;
        default:
@@ -2080,9 +750,8 @@ iscsi_conn_set_param(struct iscsi_cls_conn *cls_conn, enum iscsi_param param,
        return 0;
 }
 
-static int
-iscsi_tcp_conn_get_param(struct iscsi_cls_conn *cls_conn,
-                        enum iscsi_param param, char *buf)
+static int iscsi_sw_tcp_conn_get_param(struct iscsi_cls_conn *cls_conn,
+                                      enum iscsi_param param, char *buf)
 {
        struct iscsi_conn *conn = cls_conn->dd_data;
        int len;
@@ -2105,129 +774,119 @@ iscsi_tcp_conn_get_param(struct iscsi_cls_conn *cls_conn,
        return len;
 }
 
-static int
-iscsi_tcp_host_get_param(struct Scsi_Host *shost, enum iscsi_host_param param,
-                        char *buf)
-{
-        struct iscsi_session *session = iscsi_hostdata(shost->hostdata);
-       int len;
-
-       switch (param) {
-       case ISCSI_HOST_PARAM_IPADDRESS:
-               spin_lock_bh(&session->lock);
-               if (!session->leadconn)
-                       len = -ENODEV;
-               else
-                       len = sprintf(buf, "%s\n",
-                                    session->leadconn->local_address);
-               spin_unlock_bh(&session->lock);
-               break;
-       default:
-               return iscsi_host_get_param(shost, param, buf);
-       }
-       return len;
-}
-
 static void
-iscsi_conn_get_stats(struct iscsi_cls_conn *cls_conn, struct iscsi_stats *stats)
+iscsi_sw_tcp_conn_get_stats(struct iscsi_cls_conn *cls_conn,
+                           struct iscsi_stats *stats)
 {
        struct iscsi_conn *conn = cls_conn->dd_data;
        struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
+       struct iscsi_sw_tcp_conn *tcp_sw_conn = tcp_conn->dd_data;
 
-       stats->txdata_octets = conn->txdata_octets;
-       stats->rxdata_octets = conn->rxdata_octets;
-       stats->scsicmd_pdus = conn->scsicmd_pdus_cnt;
-       stats->dataout_pdus = conn->dataout_pdus_cnt;
-       stats->scsirsp_pdus = conn->scsirsp_pdus_cnt;
-       stats->datain_pdus = conn->datain_pdus_cnt;
-       stats->r2t_pdus = conn->r2t_pdus_cnt;
-       stats->tmfcmd_pdus = conn->tmfcmd_pdus_cnt;
-       stats->tmfrsp_pdus = conn->tmfrsp_pdus_cnt;
        stats->custom_length = 3;
        strcpy(stats->custom[0].desc, "tx_sendpage_failures");
-       stats->custom[0].value = tcp_conn->sendpage_failures_cnt;
+       stats->custom[0].value = tcp_sw_conn->sendpage_failures_cnt;
        strcpy(stats->custom[1].desc, "rx_discontiguous_hdr");
-       stats->custom[1].value = tcp_conn->discontiguous_hdr_cnt;
+       stats->custom[1].value = tcp_sw_conn->discontiguous_hdr_cnt;
        strcpy(stats->custom[2].desc, "eh_abort_cnt");
        stats->custom[2].value = conn->eh_abort_cnt;
+
+       iscsi_tcp_conn_get_stats(cls_conn, stats);
 }
 
 static struct iscsi_cls_session *
-iscsi_tcp_session_create(struct iscsi_transport *iscsit,
-                        struct scsi_transport_template *scsit,
-                        uint16_t cmds_max, uint16_t qdepth,
-                        uint32_t initial_cmdsn, uint32_t *hostno)
+iscsi_sw_tcp_session_create(struct iscsi_endpoint *ep, uint16_t cmds_max,
+                           uint16_t qdepth, uint32_t initial_cmdsn)
 {
        struct iscsi_cls_session *cls_session;
        struct iscsi_session *session;
-       uint32_t hn;
-       int cmd_i;
+       struct Scsi_Host *shost;
 
-       cls_session = iscsi_session_setup(iscsit, scsit, cmds_max, qdepth,
-                                        sizeof(struct iscsi_tcp_cmd_task),
-                                        sizeof(struct iscsi_tcp_mgmt_task),
-                                        initial_cmdsn, &hn);
-       if (!cls_session)
+       if (ep) {
+               printk(KERN_ERR "iscsi_tcp: invalid ep %p.\n", ep);
                return NULL;
-       *hostno = hn;
-
-       session = class_to_transport_session(cls_session);
-       for (cmd_i = 0; cmd_i < session->cmds_max; cmd_i++) {
-               struct iscsi_cmd_task *ctask = session->cmds[cmd_i];
-               struct iscsi_tcp_cmd_task *tcp_ctask = ctask->dd_data;
-
-               ctask->hdr = &tcp_ctask->hdr;
-       }
-
-       for (cmd_i = 0; cmd_i < session->mgmtpool_max; cmd_i++) {
-               struct iscsi_mgmt_task *mtask = session->mgmt_cmds[cmd_i];
-               struct iscsi_tcp_mgmt_task *tcp_mtask = mtask->dd_data;
-
-               mtask->hdr = &tcp_mtask->hdr;
        }
 
-       if (iscsi_r2tpool_alloc(class_to_transport_session(cls_session)))
-               goto r2tpool_alloc_fail;
+       shost = iscsi_host_alloc(&iscsi_sw_tcp_sht, 0, 1);
+       if (!shost)
+               return NULL;
+       shost->transportt = iscsi_sw_tcp_scsi_transport;
+       shost->cmd_per_lun = qdepth;
+       shost->max_lun = iscsi_max_lun;
+       shost->max_id = 0;
+       shost->max_channel = 0;
+       shost->max_cmd_len = SCSI_MAX_VARLEN_CDB_SIZE;
+
+       if (iscsi_host_add(shost, NULL))
+               goto free_host;
+
+       cls_session = iscsi_session_setup(&iscsi_sw_tcp_transport, shost,
+                                         cmds_max, 0,
+                                         sizeof(struct iscsi_tcp_task) +
+                                         sizeof(struct iscsi_sw_tcp_hdrbuf),
+                                         initial_cmdsn, 0);
+       if (!cls_session)
+               goto remove_host;
+       session = cls_session->dd_data;
 
+       shost->can_queue = session->scsi_cmds_max;
+       if (iscsi_tcp_r2tpool_alloc(session))
+               goto remove_session;
        return cls_session;
 
-r2tpool_alloc_fail:
+remove_session:
        iscsi_session_teardown(cls_session);
+remove_host:
+       iscsi_host_remove(shost);
+free_host:
+       iscsi_host_free(shost);
        return NULL;
 }
 
-static void iscsi_tcp_session_destroy(struct iscsi_cls_session *cls_session)
+static void iscsi_sw_tcp_session_destroy(struct iscsi_cls_session *cls_session)
 {
-       iscsi_r2tpool_free(class_to_transport_session(cls_session));
+       struct Scsi_Host *shost = iscsi_session_to_shost(cls_session);
+
+       iscsi_tcp_r2tpool_free(cls_session->dd_data);
        iscsi_session_teardown(cls_session);
+
+       iscsi_host_remove(shost);
+       iscsi_host_free(shost);
+}
+
+static int iscsi_sw_tcp_slave_alloc(struct scsi_device *sdev)
+{
+       set_bit(QUEUE_FLAG_BIDI, &sdev->request_queue->queue_flags);
+       return 0;
 }
 
-static int iscsi_tcp_slave_configure(struct scsi_device *sdev)
+static int iscsi_sw_tcp_slave_configure(struct scsi_device *sdev)
 {
        blk_queue_bounce_limit(sdev->request_queue, BLK_BOUNCE_ANY);
        blk_queue_dma_alignment(sdev->request_queue, 0);
        return 0;
 }
 
-static struct scsi_host_template iscsi_sht = {
+static struct scsi_host_template iscsi_sw_tcp_sht = {
        .module                 = THIS_MODULE,
        .name                   = "iSCSI Initiator over TCP/IP",
        .queuecommand           = iscsi_queuecommand,
        .change_queue_depth     = iscsi_change_queue_depth,
        .can_queue              = ISCSI_DEF_XMIT_CMDS_MAX - 1,
-       .sg_tablesize           = ISCSI_SG_TABLESIZE,
+       .sg_tablesize           = 4096,
        .max_sectors            = 0xFFFF,
        .cmd_per_lun            = ISCSI_DEF_CMD_PER_LUN,
        .eh_abort_handler       = iscsi_eh_abort,
        .eh_device_reset_handler= iscsi_eh_device_reset,
-       .eh_host_reset_handler  = iscsi_eh_host_reset,
+       .eh_target_reset_handler = iscsi_eh_recover_target,
        .use_clustering         = DISABLE_CLUSTERING,
-       .slave_configure        = iscsi_tcp_slave_configure,
+       .slave_alloc            = iscsi_sw_tcp_slave_alloc,
+       .slave_configure        = iscsi_sw_tcp_slave_configure,
+       .target_alloc           = iscsi_target_alloc,
        .proc_name              = "iscsi_tcp",
        .this_id                = -1,
 };
 
-static struct iscsi_transport iscsi_tcp_transport = {
+static struct iscsi_transport iscsi_sw_tcp_transport = {
        .owner                  = THIS_MODULE,
        .name                   = "tcp",
        .caps                   = CAP_RECOVERY_L0 | CAP_MULTI_R2T | CAP_HDRDGST
@@ -2252,62 +911,63 @@ static struct iscsi_transport iscsi_tcp_transport = {
                                  ISCSI_TARGET_NAME | ISCSI_TPGT |
                                  ISCSI_USERNAME | ISCSI_PASSWORD |
                                  ISCSI_USERNAME_IN | ISCSI_PASSWORD_IN |
-                                 ISCSI_FAST_ABORT,
+                                 ISCSI_FAST_ABORT | ISCSI_ABORT_TMO |
+                                 ISCSI_LU_RESET_TMO | ISCSI_TGT_RESET_TMO |
+                                 ISCSI_PING_TMO | ISCSI_RECV_TMO |
+                                 ISCSI_IFACE_NAME | ISCSI_INITIATOR_NAME,
        .host_param_mask        = ISCSI_HOST_HWADDRESS | ISCSI_HOST_IPADDRESS |
                                  ISCSI_HOST_INITIATOR_NAME |
                                  ISCSI_HOST_NETDEV_NAME,
-       .host_template          = &iscsi_sht,
-       .conndata_size          = sizeof(struct iscsi_conn),
-       .max_conn               = 1,
-       .max_cmd_len            = ISCSI_TCP_MAX_CMD_LEN,
        /* session management */
-       .create_session         = iscsi_tcp_session_create,
-       .destroy_session        = iscsi_tcp_session_destroy,
+       .create_session         = iscsi_sw_tcp_session_create,
+       .destroy_session        = iscsi_sw_tcp_session_destroy,
        /* connection management */
-       .create_conn            = iscsi_tcp_conn_create,
-       .bind_conn              = iscsi_tcp_conn_bind,
-       .destroy_conn           = iscsi_tcp_conn_destroy,
-       .set_param              = iscsi_conn_set_param,
-       .get_conn_param         = iscsi_tcp_conn_get_param,
+       .create_conn            = iscsi_sw_tcp_conn_create,
+       .bind_conn              = iscsi_sw_tcp_conn_bind,
+       .destroy_conn           = iscsi_sw_tcp_conn_destroy,
+       .set_param              = iscsi_sw_tcp_conn_set_param,
+       .get_conn_param         = iscsi_sw_tcp_conn_get_param,
        .get_session_param      = iscsi_session_get_param,
        .start_conn             = iscsi_conn_start,
-       .stop_conn              = iscsi_tcp_conn_stop,
+       .stop_conn              = iscsi_sw_tcp_conn_stop,
        /* iscsi host params */
-       .get_host_param         = iscsi_tcp_host_get_param,
+       .get_host_param         = iscsi_host_get_param,
        .set_host_param         = iscsi_host_set_param,
        /* IO */
        .send_pdu               = iscsi_conn_send_pdu,
-       .get_stats              = iscsi_conn_get_stats,
-       .init_cmd_task          = iscsi_tcp_cmd_init,
-       .init_mgmt_task         = iscsi_tcp_mgmt_init,
-       .xmit_cmd_task          = iscsi_tcp_ctask_xmit,
-       .xmit_mgmt_task         = iscsi_tcp_mtask_xmit,
-       .cleanup_cmd_task       = iscsi_tcp_cleanup_ctask,
+       .get_stats              = iscsi_sw_tcp_conn_get_stats,
+       /* iscsi task/cmd helpers */
+       .init_task              = iscsi_tcp_task_init,
+       .xmit_task              = iscsi_tcp_task_xmit,
+       .cleanup_task           = iscsi_tcp_cleanup_task,
+       /* low level pdu helpers */
+       .xmit_pdu               = iscsi_sw_tcp_pdu_xmit,
+       .init_pdu               = iscsi_sw_tcp_pdu_init,
+       .alloc_pdu              = iscsi_sw_tcp_pdu_alloc,
        /* recovery */
        .session_recovery_timedout = iscsi_session_recovery_timedout,
 };
 
-static int __init
-iscsi_tcp_init(void)
+static int __init iscsi_sw_tcp_init(void)
 {
        if (iscsi_max_lun < 1) {
                printk(KERN_ERR "iscsi_tcp: Invalid max_lun value of %u\n",
                       iscsi_max_lun);
                return -EINVAL;
        }
-       iscsi_tcp_transport.max_lun = iscsi_max_lun;
 
-       if (!iscsi_register_transport(&iscsi_tcp_transport))
+       iscsi_sw_tcp_scsi_transport = iscsi_register_transport(
+                                               &iscsi_sw_tcp_transport);
+       if (!iscsi_sw_tcp_scsi_transport)
                return -ENODEV;
 
        return 0;
 }
 
-static void __exit
-iscsi_tcp_exit(void)
+static void __exit iscsi_sw_tcp_exit(void)
 {
-       iscsi_unregister_transport(&iscsi_tcp_transport);
+       iscsi_unregister_transport(&iscsi_sw_tcp_transport);
 }
 
-module_init(iscsi_tcp_init);
-module_exit(iscsi_tcp_exit);
+module_init(iscsi_sw_tcp_init);
+module_exit(iscsi_sw_tcp_exit);