2 * Copyright (c) 2004, 2005, 2006 Voltaire, Inc. All rights reserved.
4 * This software is available to you under a choice of one of two
5 * licenses. You may choose to be licensed under the terms of the GNU
6 * General Public License (GPL) Version 2, available from the file
7 * COPYING in the main directory of this source tree, or the
8 * OpenIB.org BSD license below:
10 * Redistribution and use in source and binary forms, with or
11 * without modification, are permitted provided that the following
14 * - Redistributions of source code must retain the above
15 * copyright notice, this list of conditions and the following
18 * - Redistributions in binary form must reproduce the above
19 * copyright notice, this list of conditions and the following
20 * disclaimer in the documentation and/or other materials
21 * provided with the distribution.
23 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
24 * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
25 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
26 * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
27 * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
28 * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
29 * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
32 #include <linux/kernel.h>
33 #include <linux/slab.h>
35 #include <linux/scatterlist.h>
36 #include <linux/kfifo.h>
37 #include <scsi/scsi_cmnd.h>
38 #include <scsi/scsi_host.h>
40 #include "iscsi_iser.h"
42 /* Constant PDU lengths calculations */
43 #define ISER_TOTAL_HEADERS_LEN (sizeof (struct iser_hdr) + \
44 sizeof (struct iscsi_hdr))
46 /* iser_dto_add_regd_buff - increments the reference count for *
47 * the registered buffer & adds it to the DTO object */
48 static void iser_dto_add_regd_buff(struct iser_dto *dto,
49 struct iser_regd_buf *regd_buf,
50 unsigned long use_offset,
51 unsigned long use_size)
55 atomic_inc(®d_buf->ref_count);
57 add_idx = dto->regd_vector_len;
58 dto->regd[add_idx] = regd_buf;
59 dto->used_sz[add_idx] = use_size;
60 dto->offset[add_idx] = use_offset;
62 dto->regd_vector_len++;
65 /* Register user buffer memory and initialize passive rdma
66 * dto descriptor. Total data size is stored in
67 * iser_task->data[ISER_DIR_IN].data_len
69 static int iser_prepare_read_cmd(struct iscsi_task *task,
73 struct iscsi_iser_task *iser_task = task->dd_data;
74 struct iser_regd_buf *regd_buf;
76 struct iser_hdr *hdr = &iser_task->desc.iser_header;
77 struct iser_data_buf *buf_in = &iser_task->data[ISER_DIR_IN];
79 err = iser_dma_map_task_data(iser_task,
86 if (edtl > iser_task->data[ISER_DIR_IN].data_len) {
87 iser_err("Total data length: %ld, less than EDTL: "
88 "%d, in READ cmd BHS itt: %d, conn: 0x%p\n",
89 iser_task->data[ISER_DIR_IN].data_len, edtl,
90 task->itt, iser_task->iser_conn);
94 err = iser_reg_rdma_mem(iser_task,ISER_DIR_IN);
96 iser_err("Failed to set up Data-IN RDMA\n");
99 regd_buf = &iser_task->rdma_regd[ISER_DIR_IN];
101 hdr->flags |= ISER_RSV;
102 hdr->read_stag = cpu_to_be32(regd_buf->reg.rkey);
103 hdr->read_va = cpu_to_be64(regd_buf->reg.va);
105 iser_dbg("Cmd itt:%d READ tags RKEY:%#.4X VA:%#llX\n",
106 task->itt, regd_buf->reg.rkey,
107 (unsigned long long)regd_buf->reg.va);
112 /* Register user buffer memory and initialize passive rdma
113 * dto descriptor. Total data size is stored in
114 * task->data[ISER_DIR_OUT].data_len
117 iser_prepare_write_cmd(struct iscsi_task *task,
119 unsigned int unsol_sz,
122 struct iscsi_iser_task *iser_task = task->dd_data;
123 struct iser_regd_buf *regd_buf;
125 struct iser_dto *send_dto = &iser_task->desc.dto;
126 struct iser_hdr *hdr = &iser_task->desc.iser_header;
127 struct iser_data_buf *buf_out = &iser_task->data[ISER_DIR_OUT];
129 err = iser_dma_map_task_data(iser_task,
136 if (edtl > iser_task->data[ISER_DIR_OUT].data_len) {
137 iser_err("Total data length: %ld, less than EDTL: %d, "
138 "in WRITE cmd BHS itt: %d, conn: 0x%p\n",
139 iser_task->data[ISER_DIR_OUT].data_len,
140 edtl, task->itt, task->conn);
144 err = iser_reg_rdma_mem(iser_task,ISER_DIR_OUT);
146 iser_err("Failed to register write cmd RDMA mem\n");
150 regd_buf = &iser_task->rdma_regd[ISER_DIR_OUT];
152 if (unsol_sz < edtl) {
153 hdr->flags |= ISER_WSV;
154 hdr->write_stag = cpu_to_be32(regd_buf->reg.rkey);
155 hdr->write_va = cpu_to_be64(regd_buf->reg.va + unsol_sz);
157 iser_dbg("Cmd itt:%d, WRITE tags, RKEY:%#.4X "
158 "VA:%#llX + unsol:%d\n",
159 task->itt, regd_buf->reg.rkey,
160 (unsigned long long)regd_buf->reg.va, unsol_sz);
164 iser_dbg("Cmd itt:%d, WRITE, adding imm.data sz: %d\n",
166 iser_dto_add_regd_buff(send_dto,
176 * iser_post_receive_control - allocates, initializes and posts receive DTO.
178 static int iser_post_receive_control(struct iscsi_conn *conn)
180 struct iscsi_iser_conn *iser_conn = conn->dd_data;
181 struct iser_desc *rx_desc;
182 struct iser_regd_buf *regd_hdr;
183 struct iser_regd_buf *regd_data;
184 struct iser_dto *recv_dto = NULL;
185 struct iser_device *device = iser_conn->ib_conn->device;
186 int rx_data_size, err = 0;
188 rx_desc = kmem_cache_alloc(ig.desc_cache, GFP_NOIO);
189 if (rx_desc == NULL) {
190 iser_err("Failed to alloc desc for post recv\n");
193 rx_desc->type = ISCSI_RX;
195 /* for the login sequence we must support rx of upto 8K; login is done
196 * after conn create/bind (connect) and conn stop/bind (reconnect),
197 * what's common for both schemes is that the connection is not started
199 if (conn->c_stage != ISCSI_CONN_STARTED)
200 rx_data_size = ISCSI_DEF_MAX_RECV_SEG_LEN;
201 else /* FIXME till user space sets conn->max_recv_dlength correctly */
204 rx_desc->data = kmalloc(rx_data_size, GFP_NOIO);
205 if (rx_desc->data == NULL) {
206 iser_err("Failed to alloc data buf for post recv\n");
208 goto post_rx_kmalloc_failure;
211 recv_dto = &rx_desc->dto;
212 recv_dto->ib_conn = iser_conn->ib_conn;
213 recv_dto->regd_vector_len = 0;
215 regd_hdr = &rx_desc->hdr_regd_buf;
216 memset(regd_hdr, 0, sizeof(struct iser_regd_buf));
217 regd_hdr->device = device;
218 regd_hdr->virt_addr = rx_desc; /* == &rx_desc->iser_header */
219 regd_hdr->data_size = ISER_TOTAL_HEADERS_LEN;
221 iser_reg_single(device, regd_hdr, DMA_FROM_DEVICE);
223 iser_dto_add_regd_buff(recv_dto, regd_hdr, 0, 0);
225 regd_data = &rx_desc->data_regd_buf;
226 memset(regd_data, 0, sizeof(struct iser_regd_buf));
227 regd_data->device = device;
228 regd_data->virt_addr = rx_desc->data;
229 regd_data->data_size = rx_data_size;
231 iser_reg_single(device, regd_data, DMA_FROM_DEVICE);
233 iser_dto_add_regd_buff(recv_dto, regd_data, 0, 0);
235 err = iser_post_recv(rx_desc);
239 /* iser_post_recv failed */
240 iser_dto_buffs_release(recv_dto);
241 kfree(rx_desc->data);
242 post_rx_kmalloc_failure:
243 kmem_cache_free(ig.desc_cache, rx_desc);
247 /* creates a new tx descriptor and adds header regd buffer */
248 static void iser_create_send_desc(struct iscsi_iser_conn *iser_conn,
249 struct iser_desc *tx_desc)
251 struct iser_regd_buf *regd_hdr = &tx_desc->hdr_regd_buf;
252 struct iser_dto *send_dto = &tx_desc->dto;
254 memset(regd_hdr, 0, sizeof(struct iser_regd_buf));
255 regd_hdr->device = iser_conn->ib_conn->device;
256 regd_hdr->virt_addr = tx_desc; /* == &tx_desc->iser_header */
257 regd_hdr->data_size = ISER_TOTAL_HEADERS_LEN;
259 send_dto->ib_conn = iser_conn->ib_conn;
260 send_dto->notify_enable = 1;
261 send_dto->regd_vector_len = 0;
263 memset(&tx_desc->iser_header, 0, sizeof(struct iser_hdr));
264 tx_desc->iser_header.flags = ISER_VER;
266 iser_dto_add_regd_buff(send_dto, regd_hdr, 0, 0);
270 * iser_conn_set_full_featured_mode - (iSER API)
272 int iser_conn_set_full_featured_mode(struct iscsi_conn *conn)
274 struct iscsi_iser_conn *iser_conn = conn->dd_data;
277 /* no need to keep it in a var, we are after login so if this should
278 * be negotiated, by now the result should be available here */
279 int initial_post_recv_bufs_num = ISER_MAX_RX_MISC_PDUS;
281 iser_dbg("Initially post: %d\n", initial_post_recv_bufs_num);
283 /* Check that there is no posted recv or send buffers left - */
284 /* they must be consumed during the login phase */
285 BUG_ON(atomic_read(&iser_conn->ib_conn->post_recv_buf_count) != 0);
286 BUG_ON(atomic_read(&iser_conn->ib_conn->post_send_buf_count) != 0);
288 /* Initial post receive buffers */
289 for (i = 0; i < initial_post_recv_bufs_num; i++) {
290 if (iser_post_receive_control(conn) != 0) {
291 iser_err("Failed to post recv bufs at:%d conn:0x%p\n",
296 iser_dbg("Posted %d post recv bufs, conn:0x%p\n", i, conn);
301 iser_check_xmit(struct iscsi_conn *conn, void *task)
303 struct iscsi_iser_conn *iser_conn = conn->dd_data;
305 if (atomic_read(&iser_conn->ib_conn->post_send_buf_count) ==
306 ISER_QP_MAX_REQ_DTOS) {
307 iser_dbg("%ld can't xmit task %p\n",jiffies,task);
315 * iser_send_command - send command PDU
317 int iser_send_command(struct iscsi_conn *conn,
318 struct iscsi_task *task)
320 struct iscsi_iser_conn *iser_conn = conn->dd_data;
321 struct iscsi_iser_task *iser_task = task->dd_data;
322 struct iser_dto *send_dto = NULL;
325 struct iser_data_buf *data_buf;
327 struct iscsi_cmd *hdr = task->hdr;
328 struct scsi_cmnd *sc = task->sc;
330 if (!iser_conn_state_comp(iser_conn->ib_conn, ISER_CONN_UP)) {
331 iser_err("Failed to send, conn: 0x%p is not up\n", iser_conn->ib_conn);
334 if (iser_check_xmit(conn, task))
337 edtl = ntohl(hdr->data_length);
339 /* build the tx desc regd header and add it to the tx desc dto */
340 iser_task->desc.type = ISCSI_TX_SCSI_COMMAND;
341 send_dto = &iser_task->desc.dto;
342 send_dto->task = iser_task;
343 iser_create_send_desc(iser_conn, &iser_task->desc);
345 if (hdr->flags & ISCSI_FLAG_CMD_READ)
346 data_buf = &iser_task->data[ISER_DIR_IN];
348 data_buf = &iser_task->data[ISER_DIR_OUT];
350 if (scsi_sg_count(sc)) { /* using a scatter list */
351 data_buf->buf = scsi_sglist(sc);
352 data_buf->size = scsi_sg_count(sc);
355 data_buf->data_len = scsi_bufflen(sc);
357 if (hdr->flags & ISCSI_FLAG_CMD_READ) {
358 err = iser_prepare_read_cmd(task, edtl);
360 goto send_command_error;
362 if (hdr->flags & ISCSI_FLAG_CMD_WRITE) {
363 err = iser_prepare_write_cmd(task,
369 goto send_command_error;
372 iser_reg_single(iser_conn->ib_conn->device,
373 send_dto->regd[0], DMA_TO_DEVICE);
375 if (iser_post_receive_control(conn) != 0) {
376 iser_err("post_recv failed!\n");
378 goto send_command_error;
381 iser_task->status = ISER_TASK_STATUS_STARTED;
383 err = iser_post_send(&iser_task->desc);
388 iser_dto_buffs_release(send_dto);
389 iser_err("conn %p failed task->itt %d err %d\n",conn, task->itt, err);
394 * iser_send_data_out - send data out PDU
396 int iser_send_data_out(struct iscsi_conn *conn,
397 struct iscsi_task *task,
398 struct iscsi_data *hdr)
400 struct iscsi_iser_conn *iser_conn = conn->dd_data;
401 struct iscsi_iser_task *iser_task = task->dd_data;
402 struct iser_desc *tx_desc = NULL;
403 struct iser_dto *send_dto = NULL;
404 unsigned long buf_offset;
405 unsigned long data_seg_len;
409 if (!iser_conn_state_comp(iser_conn->ib_conn, ISER_CONN_UP)) {
410 iser_err("Failed to send, conn: 0x%p is not up\n", iser_conn->ib_conn);
414 if (iser_check_xmit(conn, task))
417 itt = (__force uint32_t)hdr->itt;
418 data_seg_len = ntoh24(hdr->dlength);
419 buf_offset = ntohl(hdr->offset);
421 iser_dbg("%s itt %d dseg_len %d offset %d\n",
422 __func__,(int)itt,(int)data_seg_len,(int)buf_offset);
424 tx_desc = kmem_cache_alloc(ig.desc_cache, GFP_NOIO);
425 if (tx_desc == NULL) {
426 iser_err("Failed to alloc desc for post dataout\n");
430 tx_desc->type = ISCSI_TX_DATAOUT;
431 memcpy(&tx_desc->iscsi_header, hdr, sizeof(struct iscsi_hdr));
433 /* build the tx desc regd header and add it to the tx desc dto */
434 send_dto = &tx_desc->dto;
435 send_dto->task = iser_task;
436 iser_create_send_desc(iser_conn, tx_desc);
438 iser_reg_single(iser_conn->ib_conn->device,
439 send_dto->regd[0], DMA_TO_DEVICE);
441 /* all data was registered for RDMA, we can use the lkey */
442 iser_dto_add_regd_buff(send_dto,
443 &iser_task->rdma_regd[ISER_DIR_OUT],
447 if (buf_offset + data_seg_len > iser_task->data[ISER_DIR_OUT].data_len) {
448 iser_err("Offset:%ld & DSL:%ld in Data-Out "
449 "inconsistent with total len:%ld, itt:%d\n",
450 buf_offset, data_seg_len,
451 iser_task->data[ISER_DIR_OUT].data_len, itt);
453 goto send_data_out_error;
455 iser_dbg("data-out itt: %d, offset: %ld, sz: %ld\n",
456 itt, buf_offset, data_seg_len);
459 err = iser_post_send(tx_desc);
464 iser_dto_buffs_release(send_dto);
465 kmem_cache_free(ig.desc_cache, tx_desc);
466 iser_err("conn %p failed err %d\n",conn, err);
470 int iser_send_control(struct iscsi_conn *conn,
471 struct iscsi_task *task)
473 struct iscsi_iser_conn *iser_conn = conn->dd_data;
474 struct iscsi_iser_task *iser_task = task->dd_data;
475 struct iser_desc *mdesc = &iser_task->desc;
476 struct iser_dto *send_dto = NULL;
477 unsigned long data_seg_len;
479 struct iser_regd_buf *regd_buf;
480 struct iser_device *device;
482 if (!iser_conn_state_comp(iser_conn->ib_conn, ISER_CONN_UP)) {
483 iser_err("Failed to send, conn: 0x%p is not up\n", iser_conn->ib_conn);
487 if (iser_check_xmit(conn, task))
490 /* build the tx desc regd header and add it to the tx desc dto */
491 mdesc->type = ISCSI_TX_CONTROL;
492 send_dto = &mdesc->dto;
493 send_dto->task = NULL;
494 iser_create_send_desc(iser_conn, mdesc);
496 device = iser_conn->ib_conn->device;
498 iser_reg_single(device, send_dto->regd[0], DMA_TO_DEVICE);
500 data_seg_len = ntoh24(task->hdr->dlength);
502 if (data_seg_len > 0) {
503 regd_buf = &mdesc->data_regd_buf;
504 memset(regd_buf, 0, sizeof(struct iser_regd_buf));
505 regd_buf->device = device;
506 regd_buf->virt_addr = task->data;
507 regd_buf->data_size = task->data_count;
508 iser_reg_single(device, regd_buf,
510 iser_dto_add_regd_buff(send_dto, regd_buf,
515 if (iser_post_receive_control(conn) != 0) {
516 iser_err("post_rcv_buff failed!\n");
518 goto send_control_error;
521 err = iser_post_send(mdesc);
526 iser_dto_buffs_release(send_dto);
527 iser_err("conn %p failed err %d\n",conn, err);
532 * iser_rcv_dto_completion - recv DTO completion
534 void iser_rcv_completion(struct iser_desc *rx_desc,
535 unsigned long dto_xfer_len)
537 struct iser_dto *dto = &rx_desc->dto;
538 struct iscsi_iser_conn *conn = dto->ib_conn->iser_conn;
539 struct iscsi_task *task;
540 struct iscsi_iser_task *iser_task;
541 struct iscsi_hdr *hdr;
542 char *rx_data = NULL;
544 unsigned char opcode;
546 hdr = &rx_desc->iscsi_header;
548 iser_dbg("op 0x%x itt 0x%x\n", hdr->opcode,hdr->itt);
550 if (dto_xfer_len > ISER_TOTAL_HEADERS_LEN) { /* we have data */
551 rx_data_len = dto_xfer_len - ISER_TOTAL_HEADERS_LEN;
552 rx_data = dto->regd[1]->virt_addr;
553 rx_data += dto->offset[1];
556 opcode = hdr->opcode & ISCSI_OPCODE_MASK;
558 if (opcode == ISCSI_OP_SCSI_CMD_RSP) {
559 spin_lock(&conn->iscsi_conn->session->lock);
560 task = iscsi_itt_to_ctask(conn->iscsi_conn, hdr->itt);
562 __iscsi_get_task(task);
563 spin_unlock(&conn->iscsi_conn->session->lock);
566 iser_err("itt can't be matched to task!!! "
567 "conn %p opcode %d itt %d\n",
568 conn->iscsi_conn, opcode, hdr->itt);
570 iser_task = task->dd_data;
571 iser_dbg("itt %d task %p\n",hdr->itt, task);
572 iser_task->status = ISER_TASK_STATUS_COMPLETED;
573 iser_task_rdma_finalize(iser_task);
574 iscsi_put_task(task);
577 iser_dto_buffs_release(dto);
579 iscsi_iser_recv(conn->iscsi_conn, hdr, rx_data, rx_data_len);
581 kfree(rx_desc->data);
582 kmem_cache_free(ig.desc_cache, rx_desc);
584 /* decrementing conn->post_recv_buf_count only --after-- freeing the *
585 * task eliminates the need to worry on tasks which are completed in *
586 * parallel to the execution of iser_conn_term. So the code that waits *
587 * for the posted rx bufs refcount to become zero handles everything */
588 atomic_dec(&conn->ib_conn->post_recv_buf_count);
591 void iser_snd_completion(struct iser_desc *tx_desc)
593 struct iser_dto *dto = &tx_desc->dto;
594 struct iser_conn *ib_conn = dto->ib_conn;
595 struct iscsi_iser_conn *iser_conn = ib_conn->iser_conn;
596 struct iscsi_conn *conn = iser_conn->iscsi_conn;
597 struct iscsi_task *task;
600 iser_dbg("Initiator, Data sent dto=0x%p\n", dto);
602 iser_dto_buffs_release(dto);
604 if (tx_desc->type == ISCSI_TX_DATAOUT)
605 kmem_cache_free(ig.desc_cache, tx_desc);
607 if (atomic_read(&iser_conn->ib_conn->post_send_buf_count) ==
608 ISER_QP_MAX_REQ_DTOS)
611 atomic_dec(&ib_conn->post_send_buf_count);
614 iser_dbg("%ld resuming tx\n",jiffies);
615 scsi_queue_work(conn->session->host, &conn->xmitwork);
618 if (tx_desc->type == ISCSI_TX_CONTROL) {
619 /* this arithmetic is legal by libiscsi dd_data allocation */
620 task = (void *) ((long)(void *)tx_desc -
621 sizeof(struct iscsi_task));
622 if (task->hdr->itt == RESERVED_ITT)
623 iscsi_put_task(task);
627 void iser_task_rdma_init(struct iscsi_iser_task *iser_task)
630 iser_task->status = ISER_TASK_STATUS_INIT;
632 iser_task->dir[ISER_DIR_IN] = 0;
633 iser_task->dir[ISER_DIR_OUT] = 0;
635 iser_task->data[ISER_DIR_IN].data_len = 0;
636 iser_task->data[ISER_DIR_OUT].data_len = 0;
638 memset(&iser_task->rdma_regd[ISER_DIR_IN], 0,
639 sizeof(struct iser_regd_buf));
640 memset(&iser_task->rdma_regd[ISER_DIR_OUT], 0,
641 sizeof(struct iser_regd_buf));
644 void iser_task_rdma_finalize(struct iscsi_iser_task *iser_task)
647 int is_rdma_aligned = 1;
648 struct iser_regd_buf *regd;
650 /* if we were reading, copy back to unaligned sglist,
651 * anyway dma_unmap and free the copy
653 if (iser_task->data_copy[ISER_DIR_IN].copy_buf != NULL) {
655 iser_finalize_rdma_unaligned_sg(iser_task, ISER_DIR_IN);
657 if (iser_task->data_copy[ISER_DIR_OUT].copy_buf != NULL) {
659 iser_finalize_rdma_unaligned_sg(iser_task, ISER_DIR_OUT);
662 if (iser_task->dir[ISER_DIR_IN]) {
663 regd = &iser_task->rdma_regd[ISER_DIR_IN];
664 deferred = iser_regd_buff_release(regd);
666 iser_err("%d references remain for BUF-IN rdma reg\n",
667 atomic_read(®d->ref_count));
671 if (iser_task->dir[ISER_DIR_OUT]) {
672 regd = &iser_task->rdma_regd[ISER_DIR_OUT];
673 deferred = iser_regd_buff_release(regd);
675 iser_err("%d references remain for BUF-OUT rdma reg\n",
676 atomic_read(®d->ref_count));
680 /* if the data was unaligned, it was already unmapped and then copied */
682 iser_dma_unmap_task_data(iser_task);
685 void iser_dto_buffs_release(struct iser_dto *dto)
689 for (i = 0; i < dto->regd_vector_len; i++)
690 iser_regd_buff_release(dto->regd[i]);