]> bbs.cooldavid.org Git - net-next-2.6.git/blame - drivers/infiniband/ulp/iser/iscsi_iser.h
IB/iser: Revert commit bba7ebb "avoid recv buffer exhaustion"
[net-next-2.6.git] / drivers / infiniband / ulp / iser / iscsi_iser.h
CommitLineData
49cd5382
OG
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.
49cd5382
OG
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/wait.h>
49#include <linux/sched.h>
50#include <linux/list.h>
51#include <linux/slab.h>
52#include <linux/dma-mapping.h>
53#include <linux/mutex.h>
54#include <linux/mempool.h>
55#include <linux/uio.h>
56
57#include <linux/socket.h>
58#include <linux/in.h>
59#include <linux/in6.h>
60
61#include <rdma/ib_verbs.h>
62#include <rdma/ib_fmr_pool.h>
63#include <rdma/rdma_cm.h>
64
65#define DRV_NAME "iser"
66#define PFX DRV_NAME ": "
67#define DRV_VER "0.1"
68#define DRV_DATE "May 7th, 2006"
69
70#define iser_dbg(fmt, arg...) \
6f735e36
ED
71 do { \
72 if (iser_debug_level > 1) \
73 printk(KERN_DEBUG PFX "%s:" fmt,\
74 __func__ , ## arg); \
75 } while (0)
76
77#define iser_warn(fmt, arg...) \
49cd5382
OG
78 do { \
79 if (iser_debug_level > 0) \
80 printk(KERN_DEBUG PFX "%s:" fmt,\
81 __func__ , ## arg); \
82 } while (0)
83
84#define iser_err(fmt, arg...) \
85 do { \
86 printk(KERN_ERR PFX "%s:" fmt, \
87 __func__ , ## arg); \
88 } while (0)
89
8dfa0876
EZ
90#define SHIFT_4K 12
91#define SIZE_4K (1UL << SHIFT_4K)
92#define MASK_4K (~(SIZE_4K-1))
93
49cd5382 94 /* support upto 512KB in one RDMA */
8dfa0876 95#define ISCSI_ISER_SG_TABLESIZE (0x80000 >> SHIFT_4K)
e28f3d5b 96#define ISER_DEF_CMD_PER_LUN 128
49cd5382
OG
97
98/* QP settings */
99/* Maximal bounds on received asynchronous PDUs */
100#define ISER_MAX_RX_MISC_PDUS 4 /* NOOP_IN(2) , ASYNC_EVENT(2) */
101
102#define ISER_MAX_TX_MISC_PDUS 6 /* NOOP_OUT(2), TEXT(1), *
103 * SCSI_TMFUNC(2), LOGOUT(1) */
104
1548271e 105#define ISER_QP_MAX_RECV_DTOS (ISCSI_DEF_XMIT_CMDS_MAX + \
49cd5382
OG
106 ISER_MAX_RX_MISC_PDUS + \
107 ISER_MAX_TX_MISC_PDUS)
108
109/* the max TX (send) WR supported by the iSER QP is defined by *
110 * max_send_wr = T * (1 + D) + C ; D is how many inflight dataouts we expect *
111 * to have at max for SCSI command. The tx posting & completion handling code *
112 * supports -EAGAIN scheme where tx is suspended till the QP has room for more *
113 * send WR. D=8 comes from 64K/8K */
114
115#define ISER_INFLIGHT_DATAOUTS 8
116
1548271e 117#define ISER_QP_MAX_REQ_DTOS (ISCSI_DEF_XMIT_CMDS_MAX * \
49cd5382
OG
118 (1 + ISER_INFLIGHT_DATAOUTS) + \
119 ISER_MAX_TX_MISC_PDUS + \
120 ISER_MAX_RX_MISC_PDUS)
121
122#define ISER_VER 0x10
123#define ISER_WSV 0x08
124#define ISER_RSV 0x04
125
126struct iser_hdr {
127 u8 flags;
128 u8 rsvd[3];
129 __be32 write_stag; /* write rkey */
130 __be64 write_va;
131 __be32 read_stag; /* read rkey */
132 __be64 read_va;
133} __attribute__((packed));
134
135
136/* Length of an object name string */
137#define ISER_OBJECT_NAME_SIZE 64
138
139enum iser_ib_conn_state {
140 ISER_CONN_INIT, /* descriptor allocd, no conn */
141 ISER_CONN_PENDING, /* in the process of being established */
142 ISER_CONN_UP, /* up and running */
143 ISER_CONN_TERMINATING, /* in the process of being terminated */
144 ISER_CONN_DOWN, /* shut down */
145 ISER_CONN_STATES_NUM
146};
147
148enum iser_task_status {
149 ISER_TASK_STATUS_INIT = 0,
150 ISER_TASK_STATUS_STARTED,
151 ISER_TASK_STATUS_COMPLETED
152};
153
154enum iser_data_dir {
155 ISER_DIR_IN = 0, /* to initiator */
156 ISER_DIR_OUT, /* from initiator */
157 ISER_DIRS_NUM
158};
159
160struct iser_data_buf {
161 void *buf; /* pointer to the sg list */
162 unsigned int size; /* num entries of this sg */
163 unsigned long data_len; /* total data len */
164 unsigned int dma_nents; /* returned by dma_map_sg */
165 char *copy_buf; /* allocated copy buf for SGs unaligned *
166 * for rdma which are copied */
167 struct scatterlist sg_single; /* SG-ified clone of a non SG SC or *
168 * unaligned SG */
169 };
170
171/* fwd declarations */
172struct iser_device;
173struct iscsi_iser_conn;
2261ec3d 174struct iscsi_iser_task;
412eeafa 175struct iscsi_endpoint;
49cd5382
OG
176
177struct iser_mem_reg {
178 u32 lkey;
179 u32 rkey;
180 u64 va;
181 u64 len;
182 void *mem_h;
d8111028 183 int is_fmr;
49cd5382
OG
184};
185
186struct iser_regd_buf {
187 struct iser_mem_reg reg; /* memory registration info */
188 void *virt_addr;
189 struct iser_device *device; /* device->device for dma_unmap */
5180311f 190 u64 dma_addr; /* if non zero, addr for dma_unmap */
49cd5382
OG
191 enum dma_data_direction direction; /* direction for dma_unmap */
192 unsigned int data_size;
193 atomic_t ref_count; /* refcount, freed when dec to 0 */
194};
195
196#define MAX_REGD_BUF_VECTOR_LEN 2
197
198struct iser_dto {
2261ec3d 199 struct iscsi_iser_task *task;
87e8df7a 200 struct iser_conn *ib_conn;
49cd5382
OG
201 int notify_enable;
202
203 /* vector of registered buffers */
204 unsigned int regd_vector_len;
205 struct iser_regd_buf *regd[MAX_REGD_BUF_VECTOR_LEN];
206
207 /* offset into the registered buffer may be specified */
208 unsigned int offset[MAX_REGD_BUF_VECTOR_LEN];
209
210 /* a smaller size may be specified, if 0, then full size is used */
211 unsigned int used_sz[MAX_REGD_BUF_VECTOR_LEN];
212};
213
214enum iser_desc_type {
215 ISCSI_RX,
216 ISCSI_TX_CONTROL ,
217 ISCSI_TX_SCSI_COMMAND,
218 ISCSI_TX_DATAOUT
219};
220
221struct iser_desc {
222 struct iser_hdr iser_header;
223 struct iscsi_hdr iscsi_header;
224 struct iser_regd_buf hdr_regd_buf;
225 void *data; /* used by RX & TX_CONTROL */
226 struct iser_regd_buf data_regd_buf; /* used by RX & TX_CONTROL */
227 enum iser_desc_type type;
228 struct iser_dto dto;
229};
230
231struct iser_device {
232 struct ib_device *ib_device;
233 struct ib_pd *pd;
234 struct ib_cq *cq;
235 struct ib_mr *mr;
236 struct tasklet_struct cq_tasklet;
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 */
412eeafa 243 struct iscsi_endpoint *ep;
49cd5382 244 enum iser_ib_conn_state state; /* rdma connection state */
b40977d9 245 atomic_t refcount;
49cd5382
OG
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 int disc_evt_flag; /* disconn event delivered */
252 wait_queue_head_t wait; /* waitq for conn/disconn */
253 atomic_t post_recv_buf_count; /* posted rx count */
254 atomic_t post_send_buf_count; /* posted tx count */
49cd5382
OG
255 char name[ISER_OBJECT_NAME_SIZE];
256 struct iser_page_vec *page_vec; /* represents SG to fmr maps*
257 * maps serialized as tx is*/
258 struct list_head conn_list; /* entry in ig conn list */
259};
260
261struct iscsi_iser_conn {
262 struct iscsi_conn *iscsi_conn;/* ptr to iscsi conn */
263 struct iser_conn *ib_conn; /* iSER IB conn */
49cd5382
OG
264};
265
2261ec3d 266struct iscsi_iser_task {
49cd5382
OG
267 struct iser_desc desc;
268 struct iscsi_iser_conn *iser_conn;
49cd5382
OG
269 enum iser_task_status status;
270 int command_sent; /* set if command sent */
271 int dir[ISER_DIRS_NUM]; /* set if dir use*/
272 struct iser_regd_buf rdma_regd[ISER_DIRS_NUM];/* regd rdma buf */
273 struct iser_data_buf data[ISER_DIRS_NUM]; /* orig. data des*/
274 struct iser_data_buf data_copy[ISER_DIRS_NUM];/* contig. copy */
275};
276
277struct iser_page_vec {
278 u64 *pages;
279 int length;
280 int offset;
281 int data_size;
282};
283
284struct iser_global {
285 struct mutex device_list_mutex;/* */
286 struct list_head device_list; /* all iSER devices */
287 struct mutex connlist_mutex;
288 struct list_head connlist; /* all iSER IB connections */
289
e54f8188 290 struct kmem_cache *desc_cache;
49cd5382
OG
291};
292
293extern struct iser_global ig;
294extern int iser_debug_level;
295
296/* allocate connection resources needed for rdma functionality */
297int iser_conn_set_full_featured_mode(struct iscsi_conn *conn);
298
2747fdb2 299int iser_send_control(struct iscsi_conn *conn,
2261ec3d 300 struct iscsi_task *task);
49cd5382 301
2747fdb2 302int iser_send_command(struct iscsi_conn *conn,
2261ec3d 303 struct iscsi_task *task);
49cd5382 304
2747fdb2 305int iser_send_data_out(struct iscsi_conn *conn,
2261ec3d 306 struct iscsi_task *task,
2747fdb2 307 struct iscsi_data *hdr);
49cd5382
OG
308
309void iscsi_iser_recv(struct iscsi_conn *conn,
310 struct iscsi_hdr *hdr,
311 char *rx_data,
312 int rx_data_len);
313
412eeafa 314void iser_conn_init(struct iser_conn *ib_conn);
49cd5382 315
b40977d9
MC
316void iser_conn_get(struct iser_conn *ib_conn);
317
318void iser_conn_put(struct iser_conn *ib_conn);
49cd5382
OG
319
320void iser_conn_terminate(struct iser_conn *ib_conn);
321
49cd5382
OG
322void iser_rcv_completion(struct iser_desc *desc,
323 unsigned long dto_xfer_len);
324
325void iser_snd_completion(struct iser_desc *desc);
326
2261ec3d 327void iser_task_rdma_init(struct iscsi_iser_task *task);
49cd5382 328
2261ec3d 329void iser_task_rdma_finalize(struct iscsi_iser_task *task);
49cd5382
OG
330
331void iser_dto_buffs_release(struct iser_dto *dto);
332
333int iser_regd_buff_release(struct iser_regd_buf *regd_buf);
334
335void iser_reg_single(struct iser_device *device,
336 struct iser_regd_buf *regd_buf,
337 enum dma_data_direction direction);
338
2261ec3d 339void iser_finalize_rdma_unaligned_sg(struct iscsi_iser_task *task,
49cd5382
OG
340 enum iser_data_dir cmd_dir);
341
2261ec3d 342int iser_reg_rdma_mem(struct iscsi_iser_task *task,
49cd5382
OG
343 enum iser_data_dir cmd_dir);
344
345int iser_connect(struct iser_conn *ib_conn,
346 struct sockaddr_in *src_addr,
347 struct sockaddr_in *dst_addr,
348 int non_blocking);
349
350int iser_reg_page_vec(struct iser_conn *ib_conn,
351 struct iser_page_vec *page_vec,
352 struct iser_mem_reg *mem_reg);
353
354void iser_unreg_mem(struct iser_mem_reg *mem_reg);
355
356int iser_post_recv(struct iser_desc *rx_desc);
357int iser_post_send(struct iser_desc *tx_desc);
358
359int iser_conn_state_comp(struct iser_conn *ib_conn,
360 enum iser_ib_conn_state comp);
74a20780 361
2261ec3d 362int iser_dma_map_task_data(struct iscsi_iser_task *iser_task,
74a20780
EZ
363 struct iser_data_buf *data,
364 enum iser_data_dir iser_dir,
365 enum dma_data_direction dma_dir);
366
2261ec3d 367void iser_dma_unmap_task_data(struct iscsi_iser_task *iser_task);
49cd5382 368#endif