Loading...
1/*
2 * iSER transport for the Open iSCSI Initiator & iSER transport internals
3 *
4 * Copyright (C) 2004 Dmitry Yusupov
5 * Copyright (C) 2004 Alex Aizman
6 * Copyright (C) 2005 Mike Christie
7 * based on code maintained by open-iscsi@googlegroups.com
8 *
9 * Copyright (c) 2004, 2005, 2006 Voltaire, Inc. All rights reserved.
10 * Copyright (c) 2005, 2006 Cisco Systems. All rights reserved.
11 * Copyright (c) 2013-2014 Mellanox Technologies. All rights reserved.
12 *
13 * This software is available to you under a choice of one of two
14 * licenses. You may choose to be licensed under the terms of the GNU
15 * General Public License (GPL) Version 2, available from the file
16 * COPYING in the main directory of this source tree, or the
17 * OpenIB.org BSD license below:
18 *
19 * Redistribution and use in source and binary forms, with or
20 * without modification, are permitted provided that the following
21 * conditions are met:
22 *
23 * - Redistributions of source code must retain the above
24 * copyright notice, this list of conditions and the following
25 * disclaimer.
26 *
27 * - Redistributions in binary form must reproduce the above
28 * copyright notice, this list of conditions and the following
29 * disclaimer in the documentation and/or other materials
30 * provided with the distribution.
31 *
32 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
33 * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
34 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
35 * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
36 * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
37 * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
38 * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
39 * SOFTWARE.
40 */
41#ifndef __ISCSI_ISER_H__
42#define __ISCSI_ISER_H__
43
44#include <linux/types.h>
45#include <linux/net.h>
46#include <linux/printk.h>
47#include <scsi/libiscsi.h>
48#include <scsi/scsi_transport_iscsi.h>
49#include <scsi/scsi_cmnd.h>
50#include <scsi/scsi_device.h>
51#include <scsi/iser.h>
52
53#include <linux/interrupt.h>
54#include <linux/wait.h>
55#include <linux/sched.h>
56#include <linux/list.h>
57#include <linux/slab.h>
58#include <linux/dma-mapping.h>
59#include <linux/mutex.h>
60#include <linux/mempool.h>
61#include <linux/uio.h>
62
63#include <linux/socket.h>
64#include <linux/in.h>
65#include <linux/in6.h>
66
67#include <rdma/ib_verbs.h>
68#include <rdma/rdma_cm.h>
69
70#define DRV_NAME "iser"
71#define PFX DRV_NAME ": "
72#define DRV_VER "1.6"
73
74#define iser_dbg(fmt, arg...) \
75 do { \
76 if (unlikely(iser_debug_level > 2)) \
77 printk(KERN_DEBUG PFX "%s: " fmt,\
78 __func__ , ## arg); \
79 } while (0)
80
81#define iser_warn(fmt, arg...) \
82 do { \
83 if (unlikely(iser_debug_level > 0)) \
84 pr_warn(PFX "%s: " fmt, \
85 __func__ , ## arg); \
86 } while (0)
87
88#define iser_info(fmt, arg...) \
89 do { \
90 if (unlikely(iser_debug_level > 1)) \
91 pr_info(PFX "%s: " fmt, \
92 __func__ , ## arg); \
93 } while (0)
94
95#define iser_err(fmt, arg...) \
96 pr_err(PFX "%s: " fmt, __func__ , ## arg)
97
98/* Default support is 512KB I/O size */
99#define ISER_DEF_MAX_SECTORS 1024
100#define ISCSI_ISER_DEF_SG_TABLESIZE \
101 ((ISER_DEF_MAX_SECTORS * SECTOR_SIZE) >> ilog2(SZ_4K))
102/* Maximum support is 16MB I/O size */
103#define ISCSI_ISER_MAX_SG_TABLESIZE ((32768 * SECTOR_SIZE) >> ilog2(SZ_4K))
104
105#define ISER_DEF_XMIT_CMDS_DEFAULT 512
106#if ISCSI_DEF_XMIT_CMDS_MAX > ISER_DEF_XMIT_CMDS_DEFAULT
107 #define ISER_DEF_XMIT_CMDS_MAX ISCSI_DEF_XMIT_CMDS_MAX
108#else
109 #define ISER_DEF_XMIT_CMDS_MAX ISER_DEF_XMIT_CMDS_DEFAULT
110#endif
111#define ISER_DEF_CMD_PER_LUN ISER_DEF_XMIT_CMDS_MAX
112
113/* QP settings */
114/* Maximal bounds on received asynchronous PDUs */
115#define ISER_MAX_RX_MISC_PDUS 4 /* NOOP_IN(2) , ASYNC_EVENT(2) */
116
117#define ISER_MAX_TX_MISC_PDUS 6 /* NOOP_OUT(2), TEXT(1), *
118 * SCSI_TMFUNC(2), LOGOUT(1) */
119
120#define ISER_QP_MAX_RECV_DTOS (ISER_DEF_XMIT_CMDS_MAX)
121
122/* the max TX (send) WR supported by the iSER QP is defined by *
123 * max_send_wr = T * (1 + D) + C ; D is how many inflight dataouts we expect *
124 * to have at max for SCSI command. The tx posting & completion handling code *
125 * supports -EAGAIN scheme where tx is suspended till the QP has room for more *
126 * send WR. D=8 comes from 64K/8K */
127
128#define ISER_INFLIGHT_DATAOUTS 8
129
130#define ISER_QP_MAX_REQ_DTOS (ISER_DEF_XMIT_CMDS_MAX * \
131 (1 + ISER_INFLIGHT_DATAOUTS) + \
132 ISER_MAX_TX_MISC_PDUS + \
133 ISER_MAX_RX_MISC_PDUS)
134
135/* Max registration work requests per command */
136#define ISER_MAX_REG_WR_PER_CMD 5
137
138/* For Signature we don't support DATAOUTs so no need to make room for them */
139#define ISER_QP_SIG_MAX_REQ_DTOS (ISER_DEF_XMIT_CMDS_MAX * \
140 (1 + ISER_MAX_REG_WR_PER_CMD) + \
141 ISER_MAX_TX_MISC_PDUS + \
142 ISER_MAX_RX_MISC_PDUS)
143
144#define ISER_GET_MAX_XMIT_CMDS(send_wr) ((send_wr \
145 - ISER_MAX_TX_MISC_PDUS \
146 - ISER_MAX_RX_MISC_PDUS) / \
147 (1 + ISER_INFLIGHT_DATAOUTS))
148
149/* Constant PDU lengths calculations */
150#define ISER_HEADERS_LEN (sizeof(struct iser_ctrl) + sizeof(struct iscsi_hdr))
151
152#define ISER_RECV_DATA_SEG_LEN 128
153#define ISER_RX_PAYLOAD_SIZE (ISER_HEADERS_LEN + ISER_RECV_DATA_SEG_LEN)
154#define ISER_RX_LOGIN_SIZE (ISER_HEADERS_LEN + ISCSI_DEF_MAX_RECV_SEG_LEN)
155
156/* Length of an object name string */
157#define ISER_OBJECT_NAME_SIZE 64
158
159enum iser_conn_state {
160 ISER_CONN_INIT, /* descriptor allocd, no conn */
161 ISER_CONN_PENDING, /* in the process of being established */
162 ISER_CONN_UP, /* up and running */
163 ISER_CONN_TERMINATING, /* in the process of being terminated */
164 ISER_CONN_DOWN, /* shut down */
165 ISER_CONN_STATES_NUM
166};
167
168enum iser_task_status {
169 ISER_TASK_STATUS_INIT = 0,
170 ISER_TASK_STATUS_STARTED,
171 ISER_TASK_STATUS_COMPLETED
172};
173
174enum iser_data_dir {
175 ISER_DIR_IN = 0, /* to initiator */
176 ISER_DIR_OUT, /* from initiator */
177 ISER_DIRS_NUM
178};
179
180/**
181 * struct iser_data_buf - iSER data buffer
182 *
183 * @sg: pointer to the sg list
184 * @size: num entries of this sg
185 * @data_len: total buffer byte len
186 * @dma_nents: returned by dma_map_sg
187 */
188struct iser_data_buf {
189 struct scatterlist *sg;
190 int size;
191 unsigned long data_len;
192 int dma_nents;
193};
194
195/* fwd declarations */
196struct iser_device;
197struct iscsi_iser_task;
198struct iscsi_endpoint;
199struct iser_reg_resources;
200
201/**
202 * struct iser_mem_reg - iSER memory registration info
203 *
204 * @sge: memory region sg element
205 * @rkey: memory region remote key
206 * @desc: pointer to fast registration context
207 */
208struct iser_mem_reg {
209 struct ib_sge sge;
210 u32 rkey;
211 struct iser_fr_desc *desc;
212};
213
214enum iser_desc_type {
215 ISCSI_TX_CONTROL ,
216 ISCSI_TX_SCSI_COMMAND,
217 ISCSI_TX_DATAOUT
218};
219
220/**
221 * struct iser_tx_desc - iSER TX descriptor
222 *
223 * @iser_header: iser header
224 * @iscsi_header: iscsi header
225 * @type: command/control/dataout
226 * @dma_addr: header buffer dma_address
227 * @tx_sg: sg[0] points to iser/iscsi headers
228 * sg[1] optionally points to either of immediate data
229 * unsolicited data-out or control
230 * @num_sge: number sges used on this TX task
231 * @cqe: completion handler
232 * @mapped: Is the task header mapped
233 * @reg_wr: registration WR
234 * @send_wr: send WR
235 * @inv_wr: invalidate WR
236 */
237struct iser_tx_desc {
238 struct iser_ctrl iser_header;
239 struct iscsi_hdr iscsi_header;
240 enum iser_desc_type type;
241 u64 dma_addr;
242 struct ib_sge tx_sg[2];
243 int num_sge;
244 struct ib_cqe cqe;
245 bool mapped;
246 struct ib_reg_wr reg_wr;
247 struct ib_send_wr send_wr;
248 struct ib_send_wr inv_wr;
249};
250
251#define ISER_RX_PAD_SIZE (256 - (ISER_RX_PAYLOAD_SIZE + \
252 sizeof(u64) + sizeof(struct ib_sge) + \
253 sizeof(struct ib_cqe)))
254/**
255 * struct iser_rx_desc - iSER RX descriptor
256 *
257 * @iser_header: iser header
258 * @iscsi_header: iscsi header
259 * @data: received data segment
260 * @dma_addr: receive buffer dma address
261 * @rx_sg: ib_sge of receive buffer
262 * @cqe: completion handler
263 * @pad: for sense data TODO: Modify to maximum sense length supported
264 */
265struct iser_rx_desc {
266 struct iser_ctrl iser_header;
267 struct iscsi_hdr iscsi_header;
268 char data[ISER_RECV_DATA_SEG_LEN];
269 u64 dma_addr;
270 struct ib_sge rx_sg;
271 struct ib_cqe cqe;
272 char pad[ISER_RX_PAD_SIZE];
273} __packed;
274
275/**
276 * struct iser_login_desc - iSER login descriptor
277 *
278 * @req: pointer to login request buffer
279 * @rsp: pointer to login response buffer
280 * @req_dma: DMA address of login request buffer
281 * @rsp_dma: DMA address of login response buffer
282 * @sge: IB sge for login post recv
283 * @cqe: completion handler
284 */
285struct iser_login_desc {
286 void *req;
287 void *rsp;
288 u64 req_dma;
289 u64 rsp_dma;
290 struct ib_sge sge;
291 struct ib_cqe cqe;
292} __packed;
293
294struct iser_conn;
295struct ib_conn;
296
297/**
298 * struct iser_device - iSER device handle
299 *
300 * @ib_device: RDMA device
301 * @pd: Protection Domain for this device
302 * @event_handler: IB events handle routine
303 * @ig_list: entry in devices list
304 * @refcount: Reference counter, dominated by open iser connections
305 */
306struct iser_device {
307 struct ib_device *ib_device;
308 struct ib_pd *pd;
309 struct ib_event_handler event_handler;
310 struct list_head ig_list;
311 int refcount;
312};
313
314/**
315 * struct iser_reg_resources - Fast registration resources
316 *
317 * @mr: memory region
318 * @sig_mr: signature memory region
319 */
320struct iser_reg_resources {
321 struct ib_mr *mr;
322 struct ib_mr *sig_mr;
323};
324
325/**
326 * struct iser_fr_desc - Fast registration descriptor
327 *
328 * @list: entry in connection fastreg pool
329 * @rsc: data buffer registration resources
330 * @sig_protected: is region protected indicator
331 * @all_list: first and last list members
332 */
333struct iser_fr_desc {
334 struct list_head list;
335 struct iser_reg_resources rsc;
336 bool sig_protected;
337 struct list_head all_list;
338};
339
340/**
341 * struct iser_fr_pool - connection fast registration pool
342 *
343 * @list: list of fastreg descriptors
344 * @lock: protects fastreg pool
345 * @size: size of the pool
346 * @all_list: first and last list members
347 */
348struct iser_fr_pool {
349 struct list_head list;
350 spinlock_t lock;
351 int size;
352 struct list_head all_list;
353};
354
355/**
356 * struct ib_conn - Infiniband related objects
357 *
358 * @cma_id: rdma_cm connection maneger handle
359 * @qp: Connection Queue-pair
360 * @cq: Connection completion queue
361 * @cq_size: The number of max outstanding completions
362 * @device: reference to iser device
363 * @fr_pool: connection fast registration pool
364 * @pi_support: Indicate device T10-PI support
365 * @reg_cqe: completion handler
366 */
367struct ib_conn {
368 struct rdma_cm_id *cma_id;
369 struct ib_qp *qp;
370 struct ib_cq *cq;
371 u32 cq_size;
372 struct iser_device *device;
373 struct iser_fr_pool fr_pool;
374 bool pi_support;
375 struct ib_cqe reg_cqe;
376};
377
378/**
379 * struct iser_conn - iSER connection context
380 *
381 * @ib_conn: connection RDMA resources
382 * @iscsi_conn: link to matching iscsi connection
383 * @ep: transport handle
384 * @state: connection logical state
385 * @qp_max_recv_dtos: maximum number of data outs, corresponds
386 * to max number of post recvs
387 * @max_cmds: maximum cmds allowed for this connection
388 * @name: connection peer portal
389 * @release_work: deferred work for release job
390 * @state_mutex: protects iser onnection state
391 * @stop_completion: conn_stop completion
392 * @ib_completion: RDMA cleanup completion
393 * @up_completion: connection establishment completed
394 * (state is ISER_CONN_UP)
395 * @conn_list: entry in ig conn list
396 * @login_desc: login descriptor
397 * @rx_descs: rx buffers array (cyclic buffer)
398 * @num_rx_descs: number of rx descriptors
399 * @scsi_sg_tablesize: scsi host sg_tablesize
400 * @pages_per_mr: maximum pages available for registration
401 * @snd_w_inv: connection uses remote invalidation
402 */
403struct iser_conn {
404 struct ib_conn ib_conn;
405 struct iscsi_conn *iscsi_conn;
406 struct iscsi_endpoint *ep;
407 enum iser_conn_state state;
408 unsigned qp_max_recv_dtos;
409 u16 max_cmds;
410 char name[ISER_OBJECT_NAME_SIZE];
411 struct work_struct release_work;
412 struct mutex state_mutex;
413 struct completion stop_completion;
414 struct completion ib_completion;
415 struct completion up_completion;
416 struct list_head conn_list;
417 struct iser_login_desc login_desc;
418 struct iser_rx_desc *rx_descs;
419 u32 num_rx_descs;
420 unsigned short scsi_sg_tablesize;
421 unsigned short pages_per_mr;
422 bool snd_w_inv;
423};
424
425/**
426 * struct iscsi_iser_task - iser task context
427 *
428 * @desc: TX descriptor
429 * @iser_conn: link to iser connection
430 * @status: current task status
431 * @sc: link to scsi command
432 * @command_sent: indicate if command was sent
433 * @dir: iser data direction
434 * @rdma_reg: task rdma registration desc
435 * @data: iser data buffer desc
436 * @prot: iser protection buffer desc
437 */
438struct iscsi_iser_task {
439 struct iser_tx_desc desc;
440 struct iser_conn *iser_conn;
441 enum iser_task_status status;
442 struct scsi_cmnd *sc;
443 int command_sent;
444 int dir[ISER_DIRS_NUM];
445 struct iser_mem_reg rdma_reg[ISER_DIRS_NUM];
446 struct iser_data_buf data[ISER_DIRS_NUM];
447 struct iser_data_buf prot[ISER_DIRS_NUM];
448};
449
450/**
451 * struct iser_global - iSER global context
452 *
453 * @device_list_mutex: protects device_list
454 * @device_list: iser devices global list
455 * @connlist_mutex: protects connlist
456 * @connlist: iser connections global list
457 * @desc_cache: kmem cache for tx dataout
458 */
459struct iser_global {
460 struct mutex device_list_mutex;
461 struct list_head device_list;
462 struct mutex connlist_mutex;
463 struct list_head connlist;
464 struct kmem_cache *desc_cache;
465};
466
467extern struct iser_global ig;
468extern int iser_debug_level;
469extern bool iser_pi_enable;
470extern unsigned int iser_max_sectors;
471extern bool iser_always_reg;
472
473int iser_send_control(struct iscsi_conn *conn,
474 struct iscsi_task *task);
475
476int iser_send_command(struct iscsi_conn *conn,
477 struct iscsi_task *task);
478
479int iser_send_data_out(struct iscsi_conn *conn,
480 struct iscsi_task *task,
481 struct iscsi_data *hdr);
482
483void iscsi_iser_recv(struct iscsi_conn *conn,
484 struct iscsi_hdr *hdr,
485 char *rx_data,
486 int rx_data_len);
487
488void iser_conn_init(struct iser_conn *iser_conn);
489
490void iser_conn_release(struct iser_conn *iser_conn);
491
492int iser_conn_terminate(struct iser_conn *iser_conn);
493
494void iser_release_work(struct work_struct *work);
495
496void iser_err_comp(struct ib_wc *wc, const char *type);
497void iser_login_rsp(struct ib_cq *cq, struct ib_wc *wc);
498void iser_task_rsp(struct ib_cq *cq, struct ib_wc *wc);
499void iser_cmd_comp(struct ib_cq *cq, struct ib_wc *wc);
500void iser_ctrl_comp(struct ib_cq *cq, struct ib_wc *wc);
501void iser_dataout_comp(struct ib_cq *cq, struct ib_wc *wc);
502void iser_reg_comp(struct ib_cq *cq, struct ib_wc *wc);
503
504void iser_task_rdma_init(struct iscsi_iser_task *task);
505
506void iser_task_rdma_finalize(struct iscsi_iser_task *task);
507
508void iser_free_rx_descriptors(struct iser_conn *iser_conn);
509
510void iser_finalize_rdma_unaligned_sg(struct iscsi_iser_task *iser_task,
511 struct iser_data_buf *mem,
512 enum iser_data_dir cmd_dir);
513
514int iser_reg_mem_fastreg(struct iscsi_iser_task *task,
515 enum iser_data_dir dir,
516 bool all_imm);
517void iser_unreg_mem_fastreg(struct iscsi_iser_task *task,
518 enum iser_data_dir dir);
519
520int iser_connect(struct iser_conn *iser_conn,
521 struct sockaddr *src_addr,
522 struct sockaddr *dst_addr,
523 int non_blocking);
524
525int iser_post_recvl(struct iser_conn *iser_conn);
526int iser_post_recvm(struct iser_conn *iser_conn,
527 struct iser_rx_desc *rx_desc);
528int iser_post_send(struct ib_conn *ib_conn, struct iser_tx_desc *tx_desc);
529
530int iser_dma_map_task_data(struct iscsi_iser_task *iser_task,
531 enum iser_data_dir iser_dir,
532 enum dma_data_direction dma_dir);
533
534void iser_dma_unmap_task_data(struct iscsi_iser_task *iser_task,
535 enum iser_data_dir iser_dir,
536 enum dma_data_direction dma_dir);
537
538int iser_initialize_task_headers(struct iscsi_task *task,
539 struct iser_tx_desc *tx_desc);
540int iser_alloc_rx_descriptors(struct iser_conn *iser_conn,
541 struct iscsi_session *session);
542int iser_alloc_fastreg_pool(struct ib_conn *ib_conn,
543 unsigned cmds_max,
544 unsigned int size);
545void iser_free_fastreg_pool(struct ib_conn *ib_conn);
546u8 iser_check_task_pi_status(struct iscsi_iser_task *iser_task,
547 enum iser_data_dir cmd_dir, sector_t *sector);
548
549static inline struct iser_conn *
550to_iser_conn(struct ib_conn *ib_conn)
551{
552 return container_of(ib_conn, struct iser_conn, ib_conn);
553}
554
555static inline struct iser_rx_desc *
556iser_rx(struct ib_cqe *cqe)
557{
558 return container_of(cqe, struct iser_rx_desc, cqe);
559}
560
561static inline struct iser_tx_desc *
562iser_tx(struct ib_cqe *cqe)
563{
564 return container_of(cqe, struct iser_tx_desc, cqe);
565}
566
567static inline struct iser_login_desc *
568iser_login(struct ib_cqe *cqe)
569{
570 return container_of(cqe, struct iser_login_desc, cqe);
571}
572
573#endif
1/*
2 * iSER transport for the Open iSCSI Initiator & iSER transport internals
3 *
4 * Copyright (C) 2004 Dmitry Yusupov
5 * Copyright (C) 2004 Alex Aizman
6 * Copyright (C) 2005 Mike Christie
7 * based on code maintained by open-iscsi@googlegroups.com
8 *
9 * Copyright (c) 2004, 2005, 2006 Voltaire, Inc. All rights reserved.
10 * Copyright (c) 2005, 2006 Cisco Systems. All rights reserved.
11 *
12 * This software is available to you under a choice of one of two
13 * licenses. You may choose to be licensed under the terms of the GNU
14 * General Public License (GPL) Version 2, available from the file
15 * COPYING in the main directory of this source tree, or the
16 * OpenIB.org BSD license below:
17 *
18 * Redistribution and use in source and binary forms, with or
19 * without modification, are permitted provided that the following
20 * conditions are met:
21 *
22 * - Redistributions of source code must retain the above
23 * copyright notice, this list of conditions and the following
24 * disclaimer.
25 *
26 * - Redistributions in binary form must reproduce the above
27 * copyright notice, this list of conditions and the following
28 * disclaimer in the documentation and/or other materials
29 * provided with the distribution.
30 *
31 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
32 * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
33 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
34 * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
35 * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
36 * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
37 * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
38 * SOFTWARE.
39 */
40#ifndef __ISCSI_ISER_H__
41#define __ISCSI_ISER_H__
42
43#include <linux/types.h>
44#include <linux/net.h>
45#include <scsi/libiscsi.h>
46#include <scsi/scsi_transport_iscsi.h>
47
48#include <linux/interrupt.h>
49#include <linux/wait.h>
50#include <linux/sched.h>
51#include <linux/list.h>
52#include <linux/slab.h>
53#include <linux/dma-mapping.h>
54#include <linux/mutex.h>
55#include <linux/mempool.h>
56#include <linux/uio.h>
57
58#include <linux/socket.h>
59#include <linux/in.h>
60#include <linux/in6.h>
61
62#include <rdma/ib_verbs.h>
63#include <rdma/ib_fmr_pool.h>
64#include <rdma/rdma_cm.h>
65
66#define DRV_NAME "iser"
67#define PFX DRV_NAME ": "
68#define DRV_VER "0.1"
69#define DRV_DATE "May 7th, 2006"
70
71#define iser_dbg(fmt, arg...) \
72 do { \
73 if (iser_debug_level > 1) \
74 printk(KERN_DEBUG PFX "%s:" fmt,\
75 __func__ , ## arg); \
76 } while (0)
77
78#define iser_warn(fmt, arg...) \
79 do { \
80 if (iser_debug_level > 0) \
81 printk(KERN_DEBUG PFX "%s:" fmt,\
82 __func__ , ## arg); \
83 } while (0)
84
85#define iser_err(fmt, arg...) \
86 do { \
87 printk(KERN_ERR PFX "%s:" fmt, \
88 __func__ , ## arg); \
89 } while (0)
90
91#define SHIFT_4K 12
92#define SIZE_4K (1ULL << SHIFT_4K)
93#define MASK_4K (~(SIZE_4K-1))
94
95 /* support up to 512KB in one RDMA */
96#define ISCSI_ISER_SG_TABLESIZE (0x80000 >> SHIFT_4K)
97#define ISER_DEF_CMD_PER_LUN 128
98
99/* QP settings */
100/* Maximal bounds on received asynchronous PDUs */
101#define ISER_MAX_RX_MISC_PDUS 4 /* NOOP_IN(2) , ASYNC_EVENT(2) */
102
103#define ISER_MAX_TX_MISC_PDUS 6 /* NOOP_OUT(2), TEXT(1), *
104 * SCSI_TMFUNC(2), LOGOUT(1) */
105
106#define ISER_QP_MAX_RECV_DTOS (ISCSI_DEF_XMIT_CMDS_MAX)
107
108#define ISER_MIN_POSTED_RX (ISCSI_DEF_XMIT_CMDS_MAX >> 2)
109
110/* the max TX (send) WR supported by the iSER QP is defined by *
111 * max_send_wr = T * (1 + D) + C ; D is how many inflight dataouts we expect *
112 * to have at max for SCSI command. The tx posting & completion handling code *
113 * supports -EAGAIN scheme where tx is suspended till the QP has room for more *
114 * send WR. D=8 comes from 64K/8K */
115
116#define ISER_INFLIGHT_DATAOUTS 8
117
118#define ISER_QP_MAX_REQ_DTOS (ISCSI_DEF_XMIT_CMDS_MAX * \
119 (1 + ISER_INFLIGHT_DATAOUTS) + \
120 ISER_MAX_TX_MISC_PDUS + \
121 ISER_MAX_RX_MISC_PDUS)
122
123#define ISER_VER 0x10
124#define ISER_WSV 0x08
125#define ISER_RSV 0x04
126
127struct iser_hdr {
128 u8 flags;
129 u8 rsvd[3];
130 __be32 write_stag; /* write rkey */
131 __be64 write_va;
132 __be32 read_stag; /* read rkey */
133 __be64 read_va;
134} __attribute__((packed));
135
136/* Constant PDU lengths calculations */
137#define ISER_HEADERS_LEN (sizeof(struct iser_hdr) + sizeof(struct iscsi_hdr))
138
139#define ISER_RECV_DATA_SEG_LEN 128
140#define ISER_RX_PAYLOAD_SIZE (ISER_HEADERS_LEN + ISER_RECV_DATA_SEG_LEN)
141#define ISER_RX_LOGIN_SIZE (ISER_HEADERS_LEN + ISCSI_DEF_MAX_RECV_SEG_LEN)
142
143/* Length of an object name string */
144#define ISER_OBJECT_NAME_SIZE 64
145
146enum iser_ib_conn_state {
147 ISER_CONN_INIT, /* descriptor allocd, no conn */
148 ISER_CONN_PENDING, /* in the process of being established */
149 ISER_CONN_UP, /* up and running */
150 ISER_CONN_TERMINATING, /* in the process of being terminated */
151 ISER_CONN_DOWN, /* shut down */
152 ISER_CONN_STATES_NUM
153};
154
155enum iser_task_status {
156 ISER_TASK_STATUS_INIT = 0,
157 ISER_TASK_STATUS_STARTED,
158 ISER_TASK_STATUS_COMPLETED
159};
160
161enum iser_data_dir {
162 ISER_DIR_IN = 0, /* to initiator */
163 ISER_DIR_OUT, /* from initiator */
164 ISER_DIRS_NUM
165};
166
167struct iser_data_buf {
168 void *buf; /* pointer to the sg list */
169 unsigned int size; /* num entries of this sg */
170 unsigned long data_len; /* total data len */
171 unsigned int dma_nents; /* returned by dma_map_sg */
172 char *copy_buf; /* allocated copy buf for SGs unaligned *
173 * for rdma which are copied */
174 struct scatterlist sg_single; /* SG-ified clone of a non SG SC or *
175 * unaligned SG */
176 };
177
178/* fwd declarations */
179struct iser_device;
180struct iscsi_iser_conn;
181struct iscsi_iser_task;
182struct iscsi_endpoint;
183
184struct iser_mem_reg {
185 u32 lkey;
186 u32 rkey;
187 u64 va;
188 u64 len;
189 void *mem_h;
190 int is_fmr;
191};
192
193struct iser_regd_buf {
194 struct iser_mem_reg reg; /* memory registration info */
195 void *virt_addr;
196 struct iser_device *device; /* device->device for dma_unmap */
197 enum dma_data_direction direction; /* direction for dma_unmap */
198 unsigned int data_size;
199};
200
201enum iser_desc_type {
202 ISCSI_TX_CONTROL ,
203 ISCSI_TX_SCSI_COMMAND,
204 ISCSI_TX_DATAOUT
205};
206
207struct iser_tx_desc {
208 struct iser_hdr iser_header;
209 struct iscsi_hdr iscsi_header;
210 enum iser_desc_type type;
211 u64 dma_addr;
212 /* sg[0] points to iser/iscsi headers, sg[1] optionally points to either
213 of immediate data, unsolicited data-out or control (login,text) */
214 struct ib_sge tx_sg[2];
215 int num_sge;
216};
217
218#define ISER_RX_PAD_SIZE (256 - (ISER_RX_PAYLOAD_SIZE + \
219 sizeof(u64) + sizeof(struct ib_sge)))
220struct iser_rx_desc {
221 struct iser_hdr iser_header;
222 struct iscsi_hdr iscsi_header;
223 char data[ISER_RECV_DATA_SEG_LEN];
224 u64 dma_addr;
225 struct ib_sge rx_sg;
226 char pad[ISER_RX_PAD_SIZE];
227} __attribute__((packed));
228
229struct iser_device {
230 struct ib_device *ib_device;
231 struct ib_pd *pd;
232 struct ib_cq *rx_cq;
233 struct ib_cq *tx_cq;
234 struct ib_mr *mr;
235 struct tasklet_struct cq_tasklet;
236 struct ib_event_handler event_handler;
237 struct list_head ig_list; /* entry in ig devices list */
238 int refcount;
239};
240
241struct iser_conn {
242 struct iscsi_iser_conn *iser_conn; /* iser conn for upcalls */
243 struct iscsi_endpoint *ep;
244 enum iser_ib_conn_state state; /* rdma connection state */
245 atomic_t refcount;
246 spinlock_t lock; /* used for state changes */
247 struct iser_device *device; /* device context */
248 struct rdma_cm_id *cma_id; /* CMA ID */
249 struct ib_qp *qp; /* QP */
250 struct ib_fmr_pool *fmr_pool; /* pool of IB FMRs */
251 wait_queue_head_t wait; /* waitq for conn/disconn */
252 int post_recv_buf_count; /* posted rx count */
253 atomic_t post_send_buf_count; /* posted tx count */
254 char name[ISER_OBJECT_NAME_SIZE];
255 struct iser_page_vec *page_vec; /* represents SG to fmr maps*
256 * maps serialized as tx is*/
257 struct list_head conn_list; /* entry in ig conn list */
258
259 char *login_buf;
260 char *login_req_buf, *login_resp_buf;
261 u64 login_req_dma, login_resp_dma;
262 unsigned int rx_desc_head;
263 struct iser_rx_desc *rx_descs;
264 struct ib_recv_wr rx_wr[ISER_MIN_POSTED_RX];
265};
266
267struct iscsi_iser_conn {
268 struct iscsi_conn *iscsi_conn;/* ptr to iscsi conn */
269 struct iser_conn *ib_conn; /* iSER IB conn */
270};
271
272struct iscsi_iser_task {
273 struct iser_tx_desc desc;
274 struct iscsi_iser_conn *iser_conn;
275 enum iser_task_status status;
276 int command_sent; /* set if command sent */
277 int dir[ISER_DIRS_NUM]; /* set if dir use*/
278 struct iser_regd_buf rdma_regd[ISER_DIRS_NUM];/* regd rdma buf */
279 struct iser_data_buf data[ISER_DIRS_NUM]; /* orig. data des*/
280 struct iser_data_buf data_copy[ISER_DIRS_NUM];/* contig. copy */
281};
282
283struct iser_page_vec {
284 u64 *pages;
285 int length;
286 int offset;
287 int data_size;
288};
289
290struct iser_global {
291 struct mutex device_list_mutex;/* */
292 struct list_head device_list; /* all iSER devices */
293 struct mutex connlist_mutex;
294 struct list_head connlist; /* all iSER IB connections */
295
296 struct kmem_cache *desc_cache;
297};
298
299extern struct iser_global ig;
300extern int iser_debug_level;
301
302/* allocate connection resources needed for rdma functionality */
303int iser_conn_set_full_featured_mode(struct iscsi_conn *conn);
304
305int iser_send_control(struct iscsi_conn *conn,
306 struct iscsi_task *task);
307
308int iser_send_command(struct iscsi_conn *conn,
309 struct iscsi_task *task);
310
311int iser_send_data_out(struct iscsi_conn *conn,
312 struct iscsi_task *task,
313 struct iscsi_data *hdr);
314
315void iscsi_iser_recv(struct iscsi_conn *conn,
316 struct iscsi_hdr *hdr,
317 char *rx_data,
318 int rx_data_len);
319
320void iser_conn_init(struct iser_conn *ib_conn);
321
322void iser_conn_get(struct iser_conn *ib_conn);
323
324int iser_conn_put(struct iser_conn *ib_conn, int destroy_cma_id_allowed);
325
326void iser_conn_terminate(struct iser_conn *ib_conn);
327
328void iser_rcv_completion(struct iser_rx_desc *desc,
329 unsigned long dto_xfer_len,
330 struct iser_conn *ib_conn);
331
332void iser_snd_completion(struct iser_tx_desc *desc, struct iser_conn *ib_conn);
333
334void iser_task_rdma_init(struct iscsi_iser_task *task);
335
336void iser_task_rdma_finalize(struct iscsi_iser_task *task);
337
338void iser_free_rx_descriptors(struct iser_conn *ib_conn);
339
340void iser_finalize_rdma_unaligned_sg(struct iscsi_iser_task *task,
341 enum iser_data_dir cmd_dir);
342
343int iser_reg_rdma_mem(struct iscsi_iser_task *task,
344 enum iser_data_dir cmd_dir);
345
346int iser_connect(struct iser_conn *ib_conn,
347 struct sockaddr_in *src_addr,
348 struct sockaddr_in *dst_addr,
349 int non_blocking);
350
351int iser_reg_page_vec(struct iser_conn *ib_conn,
352 struct iser_page_vec *page_vec,
353 struct iser_mem_reg *mem_reg);
354
355void iser_unreg_mem(struct iser_mem_reg *mem_reg);
356
357int iser_post_recvl(struct iser_conn *ib_conn);
358int iser_post_recvm(struct iser_conn *ib_conn, int count);
359int iser_post_send(struct iser_conn *ib_conn, struct iser_tx_desc *tx_desc);
360
361int iser_dma_map_task_data(struct iscsi_iser_task *iser_task,
362 struct iser_data_buf *data,
363 enum iser_data_dir iser_dir,
364 enum dma_data_direction dma_dir);
365
366void iser_dma_unmap_task_data(struct iscsi_iser_task *iser_task);
367int iser_initialize_task_headers(struct iscsi_task *task,
368 struct iser_tx_desc *tx_desc);
369int iser_alloc_rx_descriptors(struct iser_conn *ib_conn);
370#endif