Commit | Line | Data |
---|---|---|
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. | |
39 | * | |
40 | * $Id: iscsi_iser.h 7051 2006-05-10 12:29:11Z ogerlitz $ | |
41 | */ | |
42 | #ifndef __ISCSI_ISER_H__ | |
43 | #define __ISCSI_ISER_H__ | |
44 | ||
45 | #include <linux/types.h> | |
46 | #include <linux/net.h> | |
47 | #include <scsi/libiscsi.h> | |
48 | #include <scsi/scsi_transport_iscsi.h> | |
49 | ||
50 | #include <linux/wait.h> | |
51 | #include <linux/sched.h> | |
52 | #include <linux/list.h> | |
53 | #include <linux/slab.h> | |
54 | #include <linux/dma-mapping.h> | |
55 | #include <linux/mutex.h> | |
56 | #include <linux/mempool.h> | |
57 | #include <linux/uio.h> | |
58 | ||
59 | #include <linux/socket.h> | |
60 | #include <linux/in.h> | |
61 | #include <linux/in6.h> | |
62 | ||
63 | #include <rdma/ib_verbs.h> | |
64 | #include <rdma/ib_fmr_pool.h> | |
65 | #include <rdma/rdma_cm.h> | |
66 | ||
67 | #define DRV_NAME "iser" | |
68 | #define PFX DRV_NAME ": " | |
69 | #define DRV_VER "0.1" | |
70 | #define DRV_DATE "May 7th, 2006" | |
71 | ||
72 | #define iser_dbg(fmt, arg...) \ | |
73 | do { \ | |
74 | if (iser_debug_level > 0) \ | |
75 | printk(KERN_DEBUG PFX "%s:" fmt,\ | |
76 | __func__ , ## arg); \ | |
77 | } while (0) | |
78 | ||
79 | #define iser_err(fmt, arg...) \ | |
80 | do { \ | |
81 | printk(KERN_ERR PFX "%s:" fmt, \ | |
82 | __func__ , ## arg); \ | |
83 | } while (0) | |
84 | ||
8dfa0876 EZ |
85 | #define SHIFT_4K 12 |
86 | #define SIZE_4K (1UL << SHIFT_4K) | |
87 | #define MASK_4K (~(SIZE_4K-1)) | |
88 | ||
49cd5382 | 89 | /* support upto 512KB in one RDMA */ |
8dfa0876 | 90 | #define ISCSI_ISER_SG_TABLESIZE (0x80000 >> SHIFT_4K) |
49cd5382 OG |
91 | #define ISCSI_ISER_MAX_LUN 256 |
92 | #define ISCSI_ISER_MAX_CMD_LEN 16 | |
93 | ||
94 | /* QP settings */ | |
95 | /* Maximal bounds on received asynchronous PDUs */ | |
96 | #define ISER_MAX_RX_MISC_PDUS 4 /* NOOP_IN(2) , ASYNC_EVENT(2) */ | |
97 | ||
98 | #define ISER_MAX_TX_MISC_PDUS 6 /* NOOP_OUT(2), TEXT(1), * | |
99 | * SCSI_TMFUNC(2), LOGOUT(1) */ | |
100 | ||
101 | #define ISER_QP_MAX_RECV_DTOS (ISCSI_XMIT_CMDS_MAX + \ | |
102 | ISER_MAX_RX_MISC_PDUS + \ | |
103 | ISER_MAX_TX_MISC_PDUS) | |
104 | ||
105 | /* the max TX (send) WR supported by the iSER QP is defined by * | |
106 | * max_send_wr = T * (1 + D) + C ; D is how many inflight dataouts we expect * | |
107 | * to have at max for SCSI command. The tx posting & completion handling code * | |
108 | * supports -EAGAIN scheme where tx is suspended till the QP has room for more * | |
109 | * send WR. D=8 comes from 64K/8K */ | |
110 | ||
111 | #define ISER_INFLIGHT_DATAOUTS 8 | |
112 | ||
113 | #define ISER_QP_MAX_REQ_DTOS (ISCSI_XMIT_CMDS_MAX * \ | |
114 | (1 + ISER_INFLIGHT_DATAOUTS) + \ | |
115 | ISER_MAX_TX_MISC_PDUS + \ | |
116 | ISER_MAX_RX_MISC_PDUS) | |
117 | ||
118 | #define ISER_VER 0x10 | |
119 | #define ISER_WSV 0x08 | |
120 | #define ISER_RSV 0x04 | |
121 | ||
122 | struct iser_hdr { | |
123 | u8 flags; | |
124 | u8 rsvd[3]; | |
125 | __be32 write_stag; /* write rkey */ | |
126 | __be64 write_va; | |
127 | __be32 read_stag; /* read rkey */ | |
128 | __be64 read_va; | |
129 | } __attribute__((packed)); | |
130 | ||
131 | ||
132 | /* Length of an object name string */ | |
133 | #define ISER_OBJECT_NAME_SIZE 64 | |
134 | ||
135 | enum iser_ib_conn_state { | |
136 | ISER_CONN_INIT, /* descriptor allocd, no conn */ | |
137 | ISER_CONN_PENDING, /* in the process of being established */ | |
138 | ISER_CONN_UP, /* up and running */ | |
139 | ISER_CONN_TERMINATING, /* in the process of being terminated */ | |
140 | ISER_CONN_DOWN, /* shut down */ | |
141 | ISER_CONN_STATES_NUM | |
142 | }; | |
143 | ||
144 | enum iser_task_status { | |
145 | ISER_TASK_STATUS_INIT = 0, | |
146 | ISER_TASK_STATUS_STARTED, | |
147 | ISER_TASK_STATUS_COMPLETED | |
148 | }; | |
149 | ||
150 | enum iser_data_dir { | |
151 | ISER_DIR_IN = 0, /* to initiator */ | |
152 | ISER_DIR_OUT, /* from initiator */ | |
153 | ISER_DIRS_NUM | |
154 | }; | |
155 | ||
156 | struct iser_data_buf { | |
157 | void *buf; /* pointer to the sg list */ | |
158 | unsigned int size; /* num entries of this sg */ | |
159 | unsigned long data_len; /* total data len */ | |
160 | unsigned int dma_nents; /* returned by dma_map_sg */ | |
161 | char *copy_buf; /* allocated copy buf for SGs unaligned * | |
162 | * for rdma which are copied */ | |
163 | struct scatterlist sg_single; /* SG-ified clone of a non SG SC or * | |
164 | * unaligned SG */ | |
165 | }; | |
166 | ||
167 | /* fwd declarations */ | |
168 | struct iser_device; | |
169 | struct iscsi_iser_conn; | |
170 | struct iscsi_iser_cmd_task; | |
171 | ||
172 | struct iser_mem_reg { | |
173 | u32 lkey; | |
174 | u32 rkey; | |
175 | u64 va; | |
176 | u64 len; | |
177 | void *mem_h; | |
d8111028 | 178 | int is_fmr; |
49cd5382 OG |
179 | }; |
180 | ||
181 | struct iser_regd_buf { | |
182 | struct iser_mem_reg reg; /* memory registration info */ | |
183 | void *virt_addr; | |
184 | struct iser_device *device; /* device->device for dma_unmap */ | |
185 | dma_addr_t dma_addr; /* if non zero, addr for dma_unmap */ | |
186 | enum dma_data_direction direction; /* direction for dma_unmap */ | |
187 | unsigned int data_size; | |
188 | atomic_t ref_count; /* refcount, freed when dec to 0 */ | |
189 | }; | |
190 | ||
191 | #define MAX_REGD_BUF_VECTOR_LEN 2 | |
192 | ||
193 | struct iser_dto { | |
194 | struct iscsi_iser_cmd_task *ctask; | |
87e8df7a | 195 | struct iser_conn *ib_conn; |
49cd5382 OG |
196 | int notify_enable; |
197 | ||
198 | /* vector of registered buffers */ | |
199 | unsigned int regd_vector_len; | |
200 | struct iser_regd_buf *regd[MAX_REGD_BUF_VECTOR_LEN]; | |
201 | ||
202 | /* offset into the registered buffer may be specified */ | |
203 | unsigned int offset[MAX_REGD_BUF_VECTOR_LEN]; | |
204 | ||
205 | /* a smaller size may be specified, if 0, then full size is used */ | |
206 | unsigned int used_sz[MAX_REGD_BUF_VECTOR_LEN]; | |
207 | }; | |
208 | ||
209 | enum iser_desc_type { | |
210 | ISCSI_RX, | |
211 | ISCSI_TX_CONTROL , | |
212 | ISCSI_TX_SCSI_COMMAND, | |
213 | ISCSI_TX_DATAOUT | |
214 | }; | |
215 | ||
216 | struct iser_desc { | |
217 | struct iser_hdr iser_header; | |
218 | struct iscsi_hdr iscsi_header; | |
219 | struct iser_regd_buf hdr_regd_buf; | |
220 | void *data; /* used by RX & TX_CONTROL */ | |
221 | struct iser_regd_buf data_regd_buf; /* used by RX & TX_CONTROL */ | |
222 | enum iser_desc_type type; | |
223 | struct iser_dto dto; | |
224 | }; | |
225 | ||
226 | struct iser_device { | |
227 | struct ib_device *ib_device; | |
228 | struct ib_pd *pd; | |
229 | struct ib_cq *cq; | |
230 | struct ib_mr *mr; | |
231 | struct tasklet_struct cq_tasklet; | |
232 | struct list_head ig_list; /* entry in ig devices list */ | |
233 | int refcount; | |
234 | }; | |
235 | ||
236 | struct iser_conn { | |
237 | struct iscsi_iser_conn *iser_conn; /* iser conn for upcalls */ | |
238 | enum iser_ib_conn_state state; /* rdma connection state */ | |
239 | spinlock_t lock; /* used for state changes */ | |
240 | struct iser_device *device; /* device context */ | |
241 | struct rdma_cm_id *cma_id; /* CMA ID */ | |
242 | struct ib_qp *qp; /* QP */ | |
243 | struct ib_fmr_pool *fmr_pool; /* pool of IB FMRs */ | |
244 | int disc_evt_flag; /* disconn event delivered */ | |
245 | wait_queue_head_t wait; /* waitq for conn/disconn */ | |
246 | atomic_t post_recv_buf_count; /* posted rx count */ | |
247 | atomic_t post_send_buf_count; /* posted tx count */ | |
248 | struct work_struct comperror_work; /* conn term sleepable ctx*/ | |
249 | char name[ISER_OBJECT_NAME_SIZE]; | |
250 | struct iser_page_vec *page_vec; /* represents SG to fmr maps* | |
251 | * maps serialized as tx is*/ | |
252 | struct list_head conn_list; /* entry in ig conn list */ | |
253 | }; | |
254 | ||
255 | struct iscsi_iser_conn { | |
256 | struct iscsi_conn *iscsi_conn;/* ptr to iscsi conn */ | |
257 | struct iser_conn *ib_conn; /* iSER IB conn */ | |
258 | ||
259 | rwlock_t lock; | |
260 | }; | |
261 | ||
262 | struct iscsi_iser_cmd_task { | |
263 | struct iser_desc desc; | |
264 | struct iscsi_iser_conn *iser_conn; | |
49cd5382 OG |
265 | enum iser_task_status status; |
266 | int command_sent; /* set if command sent */ | |
267 | int dir[ISER_DIRS_NUM]; /* set if dir use*/ | |
268 | struct iser_regd_buf rdma_regd[ISER_DIRS_NUM];/* regd rdma buf */ | |
269 | struct iser_data_buf data[ISER_DIRS_NUM]; /* orig. data des*/ | |
270 | struct iser_data_buf data_copy[ISER_DIRS_NUM];/* contig. copy */ | |
271 | }; | |
272 | ||
273 | struct iser_page_vec { | |
274 | u64 *pages; | |
275 | int length; | |
276 | int offset; | |
277 | int data_size; | |
278 | }; | |
279 | ||
280 | struct iser_global { | |
281 | struct mutex device_list_mutex;/* */ | |
282 | struct list_head device_list; /* all iSER devices */ | |
283 | struct mutex connlist_mutex; | |
284 | struct list_head connlist; /* all iSER IB connections */ | |
285 | ||
e54f8188 | 286 | struct kmem_cache *desc_cache; |
49cd5382 OG |
287 | }; |
288 | ||
289 | extern struct iser_global ig; | |
290 | extern int iser_debug_level; | |
291 | ||
292 | /* allocate connection resources needed for rdma functionality */ | |
293 | int iser_conn_set_full_featured_mode(struct iscsi_conn *conn); | |
294 | ||
295 | int iser_send_control(struct iscsi_conn *conn, | |
296 | struct iscsi_mgmt_task *mtask); | |
297 | ||
298 | int iser_send_command(struct iscsi_conn *conn, | |
299 | struct iscsi_cmd_task *ctask); | |
300 | ||
301 | int iser_send_data_out(struct iscsi_conn *conn, | |
302 | struct iscsi_cmd_task *ctask, | |
303 | struct iscsi_data *hdr); | |
304 | ||
305 | void iscsi_iser_recv(struct iscsi_conn *conn, | |
306 | struct iscsi_hdr *hdr, | |
307 | char *rx_data, | |
308 | int rx_data_len); | |
309 | ||
310 | int iser_conn_init(struct iser_conn **ib_conn); | |
311 | ||
312 | void iser_conn_terminate(struct iser_conn *ib_conn); | |
313 | ||
314 | void iser_conn_release(struct iser_conn *ib_conn); | |
315 | ||
316 | void iser_rcv_completion(struct iser_desc *desc, | |
317 | unsigned long dto_xfer_len); | |
318 | ||
319 | void iser_snd_completion(struct iser_desc *desc); | |
320 | ||
321 | void iser_ctask_rdma_init(struct iscsi_iser_cmd_task *ctask); | |
322 | ||
323 | void iser_ctask_rdma_finalize(struct iscsi_iser_cmd_task *ctask); | |
324 | ||
325 | void iser_dto_buffs_release(struct iser_dto *dto); | |
326 | ||
327 | int iser_regd_buff_release(struct iser_regd_buf *regd_buf); | |
328 | ||
329 | void iser_reg_single(struct iser_device *device, | |
330 | struct iser_regd_buf *regd_buf, | |
331 | enum dma_data_direction direction); | |
332 | ||
333 | int iser_start_rdma_unaligned_sg(struct iscsi_iser_cmd_task *ctask, | |
334 | enum iser_data_dir cmd_dir); | |
335 | ||
336 | void iser_finalize_rdma_unaligned_sg(struct iscsi_iser_cmd_task *ctask, | |
337 | enum iser_data_dir cmd_dir); | |
338 | ||
339 | int iser_reg_rdma_mem(struct iscsi_iser_cmd_task *ctask, | |
340 | enum iser_data_dir cmd_dir); | |
341 | ||
342 | int iser_connect(struct iser_conn *ib_conn, | |
343 | struct sockaddr_in *src_addr, | |
344 | struct sockaddr_in *dst_addr, | |
345 | int non_blocking); | |
346 | ||
347 | int iser_reg_page_vec(struct iser_conn *ib_conn, | |
348 | struct iser_page_vec *page_vec, | |
349 | struct iser_mem_reg *mem_reg); | |
350 | ||
351 | void iser_unreg_mem(struct iser_mem_reg *mem_reg); | |
352 | ||
353 | int iser_post_recv(struct iser_desc *rx_desc); | |
354 | int iser_post_send(struct iser_desc *tx_desc); | |
355 | ||
356 | int iser_conn_state_comp(struct iser_conn *ib_conn, | |
357 | enum iser_ib_conn_state comp); | |
74a20780 EZ |
358 | |
359 | int iser_dma_map_task_data(struct iscsi_iser_cmd_task *iser_ctask, | |
360 | struct iser_data_buf *data, | |
361 | enum iser_data_dir iser_dir, | |
362 | enum dma_data_direction dma_dir); | |
363 | ||
364 | void iser_dma_unmap_task_data(struct iscsi_iser_cmd_task *iser_ctask); | |
49cd5382 | 365 | #endif |