iscsi_iser.h 12 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371
  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. #include <linux/types.h>
  43. #include <linux/net.h>
  44. #include <scsi/libiscsi.h>
  45. #include <scsi/scsi_transport_iscsi.h>
  46. #include <linux/interrupt.h>
  47. #include <linux/wait.h>
  48. #include <linux/sched.h>
  49. #include <linux/list.h>
  50. #include <linux/slab.h>
  51. #include <linux/dma-mapping.h>
  52. #include <linux/mutex.h>
  53. #include <linux/mempool.h>
  54. #include <linux/uio.h>
  55. #include <linux/socket.h>
  56. #include <linux/in.h>
  57. #include <linux/in6.h>
  58. #include <rdma/ib_verbs.h>
  59. #include <rdma/ib_fmr_pool.h>
  60. #include <rdma/rdma_cm.h>
  61. #define DRV_NAME "iser"
  62. #define PFX DRV_NAME ": "
  63. #define DRV_VER "0.1"
  64. #define DRV_DATE "May 7th, 2006"
  65. #define iser_dbg(fmt, arg...) \
  66. do { \
  67. if (iser_debug_level > 1) \
  68. printk(KERN_DEBUG PFX "%s:" fmt,\
  69. __func__ , ## arg); \
  70. } while (0)
  71. #define iser_warn(fmt, arg...) \
  72. do { \
  73. if (iser_debug_level > 0) \
  74. printk(KERN_DEBUG PFX "%s:" fmt,\
  75. __func__ , ## arg); \
  76. } while (0)
  77. #define iser_err(fmt, arg...) \
  78. do { \
  79. printk(KERN_ERR PFX "%s:" fmt, \
  80. __func__ , ## arg); \
  81. } while (0)
  82. #define SHIFT_4K 12
  83. #define SIZE_4K (1ULL << SHIFT_4K)
  84. #define MASK_4K (~(SIZE_4K-1))
  85. /* support up to 512KB in one RDMA */
  86. #define ISCSI_ISER_SG_TABLESIZE (0x80000 >> SHIFT_4K)
  87. #define ISER_DEF_CMD_PER_LUN 128
  88. /* QP settings */
  89. /* Maximal bounds on received asynchronous PDUs */
  90. #define ISER_MAX_RX_MISC_PDUS 4 /* NOOP_IN(2) , ASYNC_EVENT(2) */
  91. #define ISER_MAX_TX_MISC_PDUS 6 /* NOOP_OUT(2), TEXT(1), *
  92. * SCSI_TMFUNC(2), LOGOUT(1) */
  93. #define ISER_QP_MAX_RECV_DTOS (ISCSI_DEF_XMIT_CMDS_MAX)
  94. #define ISER_MIN_POSTED_RX (ISCSI_DEF_XMIT_CMDS_MAX >> 2)
  95. /* the max TX (send) WR supported by the iSER QP is defined by *
  96. * max_send_wr = T * (1 + D) + C ; D is how many inflight dataouts we expect *
  97. * to have at max for SCSI command. The tx posting & completion handling code *
  98. * supports -EAGAIN scheme where tx is suspended till the QP has room for more *
  99. * send WR. D=8 comes from 64K/8K */
  100. #define ISER_INFLIGHT_DATAOUTS 8
  101. #define ISER_QP_MAX_REQ_DTOS (ISCSI_DEF_XMIT_CMDS_MAX * \
  102. (1 + ISER_INFLIGHT_DATAOUTS) + \
  103. ISER_MAX_TX_MISC_PDUS + \
  104. ISER_MAX_RX_MISC_PDUS)
  105. #define ISER_VER 0x10
  106. #define ISER_WSV 0x08
  107. #define ISER_RSV 0x04
  108. struct iser_hdr {
  109. u8 flags;
  110. u8 rsvd[3];
  111. __be32 write_stag; /* write rkey */
  112. __be64 write_va;
  113. __be32 read_stag; /* read rkey */
  114. __be64 read_va;
  115. } __attribute__((packed));
  116. /* Constant PDU lengths calculations */
  117. #define ISER_HEADERS_LEN (sizeof(struct iser_hdr) + sizeof(struct iscsi_hdr))
  118. #define ISER_RECV_DATA_SEG_LEN 128
  119. #define ISER_RX_PAYLOAD_SIZE (ISER_HEADERS_LEN + ISER_RECV_DATA_SEG_LEN)
  120. #define ISER_RX_LOGIN_SIZE (ISER_HEADERS_LEN + ISCSI_DEF_MAX_RECV_SEG_LEN)
  121. /* Length of an object name string */
  122. #define ISER_OBJECT_NAME_SIZE 64
  123. enum iser_ib_conn_state {
  124. ISER_CONN_INIT, /* descriptor allocd, no conn */
  125. ISER_CONN_PENDING, /* in the process of being established */
  126. ISER_CONN_UP, /* up and running */
  127. ISER_CONN_TERMINATING, /* in the process of being terminated */
  128. ISER_CONN_DOWN, /* shut down */
  129. ISER_CONN_STATES_NUM
  130. };
  131. enum iser_task_status {
  132. ISER_TASK_STATUS_INIT = 0,
  133. ISER_TASK_STATUS_STARTED,
  134. ISER_TASK_STATUS_COMPLETED
  135. };
  136. enum iser_data_dir {
  137. ISER_DIR_IN = 0, /* to initiator */
  138. ISER_DIR_OUT, /* from initiator */
  139. ISER_DIRS_NUM
  140. };
  141. struct iser_data_buf {
  142. void *buf; /* pointer to the sg list */
  143. unsigned int size; /* num entries of this sg */
  144. unsigned long data_len; /* total data len */
  145. unsigned int dma_nents; /* returned by dma_map_sg */
  146. char *copy_buf; /* allocated copy buf for SGs unaligned *
  147. * for rdma which are copied */
  148. struct scatterlist sg_single; /* SG-ified clone of a non SG SC or *
  149. * unaligned SG */
  150. };
  151. /* fwd declarations */
  152. struct iser_device;
  153. struct iscsi_iser_conn;
  154. struct iscsi_iser_task;
  155. struct iscsi_endpoint;
  156. struct iser_mem_reg {
  157. u32 lkey;
  158. u32 rkey;
  159. u64 va;
  160. u64 len;
  161. void *mem_h;
  162. int is_fmr;
  163. };
  164. struct iser_regd_buf {
  165. struct iser_mem_reg reg; /* memory registration info */
  166. void *virt_addr;
  167. struct iser_device *device; /* device->device for dma_unmap */
  168. enum dma_data_direction direction; /* direction for dma_unmap */
  169. unsigned int data_size;
  170. };
  171. enum iser_desc_type {
  172. ISCSI_TX_CONTROL ,
  173. ISCSI_TX_SCSI_COMMAND,
  174. ISCSI_TX_DATAOUT
  175. };
  176. struct iser_tx_desc {
  177. struct iser_hdr iser_header;
  178. struct iscsi_hdr iscsi_header;
  179. enum iser_desc_type type;
  180. u64 dma_addr;
  181. /* sg[0] points to iser/iscsi headers, sg[1] optionally points to either
  182. of immediate data, unsolicited data-out or control (login,text) */
  183. struct ib_sge tx_sg[2];
  184. int num_sge;
  185. };
  186. #define ISER_RX_PAD_SIZE (256 - (ISER_RX_PAYLOAD_SIZE + \
  187. sizeof(u64) + sizeof(struct ib_sge)))
  188. struct iser_rx_desc {
  189. struct iser_hdr iser_header;
  190. struct iscsi_hdr iscsi_header;
  191. char data[ISER_RECV_DATA_SEG_LEN];
  192. u64 dma_addr;
  193. struct ib_sge rx_sg;
  194. char pad[ISER_RX_PAD_SIZE];
  195. } __attribute__((packed));
  196. struct iser_device {
  197. struct ib_device *ib_device;
  198. struct ib_pd *pd;
  199. struct ib_cq *rx_cq;
  200. struct ib_cq *tx_cq;
  201. struct ib_mr *mr;
  202. struct tasklet_struct cq_tasklet;
  203. struct ib_event_handler event_handler;
  204. struct list_head ig_list; /* entry in ig devices list */
  205. int refcount;
  206. };
  207. struct iser_conn {
  208. struct iscsi_iser_conn *iser_conn; /* iser conn for upcalls */
  209. struct iscsi_endpoint *ep;
  210. enum iser_ib_conn_state state; /* rdma connection state */
  211. atomic_t refcount;
  212. spinlock_t lock; /* used for state changes */
  213. struct iser_device *device; /* device context */
  214. struct rdma_cm_id *cma_id; /* CMA ID */
  215. struct ib_qp *qp; /* QP */
  216. struct ib_fmr_pool *fmr_pool; /* pool of IB FMRs */
  217. wait_queue_head_t wait; /* waitq for conn/disconn */
  218. int post_recv_buf_count; /* posted rx count */
  219. atomic_t post_send_buf_count; /* posted tx count */
  220. char name[ISER_OBJECT_NAME_SIZE];
  221. struct iser_page_vec *page_vec; /* represents SG to fmr maps*
  222. * maps serialized as tx is*/
  223. struct list_head conn_list; /* entry in ig conn list */
  224. char *login_buf;
  225. char *login_req_buf, *login_resp_buf;
  226. u64 login_req_dma, login_resp_dma;
  227. unsigned int rx_desc_head;
  228. struct iser_rx_desc *rx_descs;
  229. struct ib_recv_wr rx_wr[ISER_MIN_POSTED_RX];
  230. };
  231. struct iscsi_iser_conn {
  232. struct iscsi_conn *iscsi_conn;/* ptr to iscsi conn */
  233. struct iser_conn *ib_conn; /* iSER IB conn */
  234. };
  235. struct iscsi_iser_task {
  236. struct iser_tx_desc desc;
  237. struct iscsi_iser_conn *iser_conn;
  238. enum iser_task_status status;
  239. int command_sent; /* set if command sent */
  240. int dir[ISER_DIRS_NUM]; /* set if dir use*/
  241. struct iser_regd_buf rdma_regd[ISER_DIRS_NUM];/* regd rdma buf */
  242. struct iser_data_buf data[ISER_DIRS_NUM]; /* orig. data des*/
  243. struct iser_data_buf data_copy[ISER_DIRS_NUM];/* contig. copy */
  244. };
  245. struct iser_page_vec {
  246. u64 *pages;
  247. int length;
  248. int offset;
  249. int data_size;
  250. };
  251. struct iser_global {
  252. struct mutex device_list_mutex;/* */
  253. struct list_head device_list; /* all iSER devices */
  254. struct mutex connlist_mutex;
  255. struct list_head connlist; /* all iSER IB connections */
  256. struct kmem_cache *desc_cache;
  257. };
  258. extern struct iser_global ig;
  259. extern int iser_debug_level;
  260. /* allocate connection resources needed for rdma functionality */
  261. int iser_conn_set_full_featured_mode(struct iscsi_conn *conn);
  262. int iser_send_control(struct iscsi_conn *conn,
  263. struct iscsi_task *task);
  264. int iser_send_command(struct iscsi_conn *conn,
  265. struct iscsi_task *task);
  266. int iser_send_data_out(struct iscsi_conn *conn,
  267. struct iscsi_task *task,
  268. struct iscsi_data *hdr);
  269. void iscsi_iser_recv(struct iscsi_conn *conn,
  270. struct iscsi_hdr *hdr,
  271. char *rx_data,
  272. int rx_data_len);
  273. void iser_conn_init(struct iser_conn *ib_conn);
  274. void iser_conn_get(struct iser_conn *ib_conn);
  275. int iser_conn_put(struct iser_conn *ib_conn, int destroy_cma_id_allowed);
  276. void iser_conn_terminate(struct iser_conn *ib_conn);
  277. void iser_rcv_completion(struct iser_rx_desc *desc,
  278. unsigned long dto_xfer_len,
  279. struct iser_conn *ib_conn);
  280. void iser_snd_completion(struct iser_tx_desc *desc, struct iser_conn *ib_conn);
  281. void iser_task_rdma_init(struct iscsi_iser_task *task);
  282. void iser_task_rdma_finalize(struct iscsi_iser_task *task);
  283. void iser_free_rx_descriptors(struct iser_conn *ib_conn);
  284. void iser_finalize_rdma_unaligned_sg(struct iscsi_iser_task *task,
  285. enum iser_data_dir cmd_dir);
  286. int iser_reg_rdma_mem(struct iscsi_iser_task *task,
  287. enum iser_data_dir cmd_dir);
  288. int iser_connect(struct iser_conn *ib_conn,
  289. struct sockaddr_in *src_addr,
  290. struct sockaddr_in *dst_addr,
  291. int non_blocking);
  292. int iser_reg_page_vec(struct iser_conn *ib_conn,
  293. struct iser_page_vec *page_vec,
  294. struct iser_mem_reg *mem_reg);
  295. void iser_unreg_mem(struct iser_mem_reg *mem_reg);
  296. int iser_post_recvl(struct iser_conn *ib_conn);
  297. int iser_post_recvm(struct iser_conn *ib_conn, int count);
  298. int iser_post_send(struct iser_conn *ib_conn, struct iser_tx_desc *tx_desc);
  299. int iser_dma_map_task_data(struct iscsi_iser_task *iser_task,
  300. struct iser_data_buf *data,
  301. enum iser_data_dir iser_dir,
  302. enum dma_data_direction dma_dir);
  303. void iser_dma_unmap_task_data(struct iscsi_iser_task *iser_task);
  304. int iser_initialize_task_headers(struct iscsi_task *task,
  305. struct iser_tx_desc *tx_desc);
  306. int iser_alloc_rx_descriptors(struct iser_conn *ib_conn);
  307. #endif