iscsi_iser.h 12 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370
  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/wait.h>
  47. #include <linux/sched.h>
  48. #include <linux/list.h>
  49. #include <linux/slab.h>
  50. #include <linux/dma-mapping.h>
  51. #include <linux/mutex.h>
  52. #include <linux/mempool.h>
  53. #include <linux/uio.h>
  54. #include <linux/socket.h>
  55. #include <linux/in.h>
  56. #include <linux/in6.h>
  57. #include <rdma/ib_verbs.h>
  58. #include <rdma/ib_fmr_pool.h>
  59. #include <rdma/rdma_cm.h>
  60. #define DRV_NAME "iser"
  61. #define PFX DRV_NAME ": "
  62. #define DRV_VER "0.1"
  63. #define DRV_DATE "May 7th, 2006"
  64. #define iser_dbg(fmt, arg...) \
  65. do { \
  66. if (iser_debug_level > 1) \
  67. printk(KERN_DEBUG PFX "%s:" fmt,\
  68. __func__ , ## arg); \
  69. } while (0)
  70. #define iser_warn(fmt, arg...) \
  71. do { \
  72. if (iser_debug_level > 0) \
  73. printk(KERN_DEBUG PFX "%s:" fmt,\
  74. __func__ , ## arg); \
  75. } while (0)
  76. #define iser_err(fmt, arg...) \
  77. do { \
  78. printk(KERN_ERR PFX "%s:" fmt, \
  79. __func__ , ## arg); \
  80. } while (0)
  81. #define SHIFT_4K 12
  82. #define SIZE_4K (1UL << SHIFT_4K)
  83. #define MASK_4K (~(SIZE_4K-1))
  84. /* support up to 512KB in one RDMA */
  85. #define ISCSI_ISER_SG_TABLESIZE (0x80000 >> SHIFT_4K)
  86. #define ISER_DEF_CMD_PER_LUN 128
  87. /* QP settings */
  88. /* Maximal bounds on received asynchronous PDUs */
  89. #define ISER_MAX_RX_MISC_PDUS 4 /* NOOP_IN(2) , ASYNC_EVENT(2) */
  90. #define ISER_MAX_TX_MISC_PDUS 6 /* NOOP_OUT(2), TEXT(1), *
  91. * SCSI_TMFUNC(2), LOGOUT(1) */
  92. #define ISER_QP_MAX_RECV_DTOS (ISCSI_DEF_XMIT_CMDS_MAX)
  93. #define ISER_MIN_POSTED_RX (ISCSI_DEF_XMIT_CMDS_MAX >> 2)
  94. /* the max TX (send) WR supported by the iSER QP is defined by *
  95. * max_send_wr = T * (1 + D) + C ; D is how many inflight dataouts we expect *
  96. * to have at max for SCSI command. The tx posting & completion handling code *
  97. * supports -EAGAIN scheme where tx is suspended till the QP has room for more *
  98. * send WR. D=8 comes from 64K/8K */
  99. #define ISER_INFLIGHT_DATAOUTS 8
  100. #define ISER_QP_MAX_REQ_DTOS (ISCSI_DEF_XMIT_CMDS_MAX * \
  101. (1 + ISER_INFLIGHT_DATAOUTS) + \
  102. ISER_MAX_TX_MISC_PDUS + \
  103. ISER_MAX_RX_MISC_PDUS)
  104. #define ISER_VER 0x10
  105. #define ISER_WSV 0x08
  106. #define ISER_RSV 0x04
  107. struct iser_hdr {
  108. u8 flags;
  109. u8 rsvd[3];
  110. __be32 write_stag; /* write rkey */
  111. __be64 write_va;
  112. __be32 read_stag; /* read rkey */
  113. __be64 read_va;
  114. } __attribute__((packed));
  115. /* Constant PDU lengths calculations */
  116. #define ISER_HEADERS_LEN (sizeof(struct iser_hdr) + sizeof(struct iscsi_hdr))
  117. #define ISER_RECV_DATA_SEG_LEN 128
  118. #define ISER_RX_PAYLOAD_SIZE (ISER_HEADERS_LEN + ISER_RECV_DATA_SEG_LEN)
  119. #define ISER_RX_LOGIN_SIZE (ISER_HEADERS_LEN + ISCSI_DEF_MAX_RECV_SEG_LEN)
  120. /* Length of an object name string */
  121. #define ISER_OBJECT_NAME_SIZE 64
  122. enum iser_ib_conn_state {
  123. ISER_CONN_INIT, /* descriptor allocd, no conn */
  124. ISER_CONN_PENDING, /* in the process of being established */
  125. ISER_CONN_UP, /* up and running */
  126. ISER_CONN_TERMINATING, /* in the process of being terminated */
  127. ISER_CONN_DOWN, /* shut down */
  128. ISER_CONN_STATES_NUM
  129. };
  130. enum iser_task_status {
  131. ISER_TASK_STATUS_INIT = 0,
  132. ISER_TASK_STATUS_STARTED,
  133. ISER_TASK_STATUS_COMPLETED
  134. };
  135. enum iser_data_dir {
  136. ISER_DIR_IN = 0, /* to initiator */
  137. ISER_DIR_OUT, /* from initiator */
  138. ISER_DIRS_NUM
  139. };
  140. struct iser_data_buf {
  141. void *buf; /* pointer to the sg list */
  142. unsigned int size; /* num entries of this sg */
  143. unsigned long data_len; /* total data len */
  144. unsigned int dma_nents; /* returned by dma_map_sg */
  145. char *copy_buf; /* allocated copy buf for SGs unaligned *
  146. * for rdma which are copied */
  147. struct scatterlist sg_single; /* SG-ified clone of a non SG SC or *
  148. * unaligned SG */
  149. };
  150. /* fwd declarations */
  151. struct iser_device;
  152. struct iscsi_iser_conn;
  153. struct iscsi_iser_task;
  154. struct iscsi_endpoint;
  155. struct iser_mem_reg {
  156. u32 lkey;
  157. u32 rkey;
  158. u64 va;
  159. u64 len;
  160. void *mem_h;
  161. int is_fmr;
  162. };
  163. struct iser_regd_buf {
  164. struct iser_mem_reg reg; /* memory registration info */
  165. void *virt_addr;
  166. struct iser_device *device; /* device->device for dma_unmap */
  167. enum dma_data_direction direction; /* direction for dma_unmap */
  168. unsigned int data_size;
  169. };
  170. enum iser_desc_type {
  171. ISCSI_TX_CONTROL ,
  172. ISCSI_TX_SCSI_COMMAND,
  173. ISCSI_TX_DATAOUT
  174. };
  175. struct iser_tx_desc {
  176. struct iser_hdr iser_header;
  177. struct iscsi_hdr iscsi_header;
  178. enum iser_desc_type type;
  179. u64 dma_addr;
  180. /* sg[0] points to iser/iscsi headers, sg[1] optionally points to either
  181. of immediate data, unsolicited data-out or control (login,text) */
  182. struct ib_sge tx_sg[2];
  183. int num_sge;
  184. };
  185. #define ISER_RX_PAD_SIZE (256 - (ISER_RX_PAYLOAD_SIZE + \
  186. sizeof(u64) + sizeof(struct ib_sge)))
  187. struct iser_rx_desc {
  188. struct iser_hdr iser_header;
  189. struct iscsi_hdr iscsi_header;
  190. char data[ISER_RECV_DATA_SEG_LEN];
  191. u64 dma_addr;
  192. struct ib_sge rx_sg;
  193. char pad[ISER_RX_PAD_SIZE];
  194. } __attribute__((packed));
  195. struct iser_device {
  196. struct ib_device *ib_device;
  197. struct ib_pd *pd;
  198. struct ib_cq *rx_cq;
  199. struct ib_cq *tx_cq;
  200. struct ib_mr *mr;
  201. struct tasklet_struct cq_tasklet;
  202. struct ib_event_handler event_handler;
  203. struct list_head ig_list; /* entry in ig devices list */
  204. int refcount;
  205. };
  206. struct iser_conn {
  207. struct iscsi_iser_conn *iser_conn; /* iser conn for upcalls */
  208. struct iscsi_endpoint *ep;
  209. enum iser_ib_conn_state state; /* rdma connection state */
  210. atomic_t refcount;
  211. spinlock_t lock; /* used for state changes */
  212. struct iser_device *device; /* device context */
  213. struct rdma_cm_id *cma_id; /* CMA ID */
  214. struct ib_qp *qp; /* QP */
  215. struct ib_fmr_pool *fmr_pool; /* pool of IB FMRs */
  216. wait_queue_head_t wait; /* waitq for conn/disconn */
  217. int post_recv_buf_count; /* posted rx count */
  218. atomic_t post_send_buf_count; /* posted tx count */
  219. char name[ISER_OBJECT_NAME_SIZE];
  220. struct iser_page_vec *page_vec; /* represents SG to fmr maps*
  221. * maps serialized as tx is*/
  222. struct list_head conn_list; /* entry in ig conn list */
  223. char *login_buf;
  224. u64 login_dma;
  225. unsigned int rx_desc_head;
  226. struct iser_rx_desc *rx_descs;
  227. struct ib_recv_wr rx_wr[ISER_MIN_POSTED_RX];
  228. };
  229. struct iscsi_iser_conn {
  230. struct iscsi_conn *iscsi_conn;/* ptr to iscsi conn */
  231. struct iser_conn *ib_conn; /* iSER IB conn */
  232. };
  233. struct iscsi_iser_task {
  234. struct iser_tx_desc desc;
  235. struct iscsi_iser_conn *iser_conn;
  236. enum iser_task_status status;
  237. int command_sent; /* set if command sent */
  238. int dir[ISER_DIRS_NUM]; /* set if dir use*/
  239. struct iser_regd_buf rdma_regd[ISER_DIRS_NUM];/* regd rdma buf */
  240. struct iser_data_buf data[ISER_DIRS_NUM]; /* orig. data des*/
  241. struct iser_data_buf data_copy[ISER_DIRS_NUM];/* contig. copy */
  242. int headers_initialized;
  243. };
  244. struct iser_page_vec {
  245. u64 *pages;
  246. int length;
  247. int offset;
  248. int data_size;
  249. };
  250. struct iser_global {
  251. struct mutex device_list_mutex;/* */
  252. struct list_head device_list; /* all iSER devices */
  253. struct mutex connlist_mutex;
  254. struct list_head connlist; /* all iSER IB connections */
  255. struct kmem_cache *desc_cache;
  256. };
  257. extern struct iser_global ig;
  258. extern int iser_debug_level;
  259. /* allocate connection resources needed for rdma functionality */
  260. int iser_conn_set_full_featured_mode(struct iscsi_conn *conn);
  261. int iser_send_control(struct iscsi_conn *conn,
  262. struct iscsi_task *task);
  263. int iser_send_command(struct iscsi_conn *conn,
  264. struct iscsi_task *task);
  265. int iser_send_data_out(struct iscsi_conn *conn,
  266. struct iscsi_task *task,
  267. struct iscsi_data *hdr);
  268. void iscsi_iser_recv(struct iscsi_conn *conn,
  269. struct iscsi_hdr *hdr,
  270. char *rx_data,
  271. int rx_data_len);
  272. void iser_conn_init(struct iser_conn *ib_conn);
  273. void iser_conn_get(struct iser_conn *ib_conn);
  274. int iser_conn_put(struct iser_conn *ib_conn, int destroy_cma_id_allowed);
  275. void iser_conn_terminate(struct iser_conn *ib_conn);
  276. void iser_rcv_completion(struct iser_rx_desc *desc,
  277. unsigned long dto_xfer_len,
  278. struct iser_conn *ib_conn);
  279. void iser_snd_completion(struct iser_tx_desc *desc, struct iser_conn *ib_conn);
  280. void iser_task_rdma_init(struct iscsi_iser_task *task);
  281. void iser_task_rdma_finalize(struct iscsi_iser_task *task);
  282. void iser_free_rx_descriptors(struct iser_conn *ib_conn);
  283. void iser_finalize_rdma_unaligned_sg(struct iscsi_iser_task *task,
  284. enum iser_data_dir cmd_dir);
  285. int iser_reg_rdma_mem(struct iscsi_iser_task *task,
  286. enum iser_data_dir cmd_dir);
  287. int iser_connect(struct iser_conn *ib_conn,
  288. struct sockaddr_in *src_addr,
  289. struct sockaddr_in *dst_addr,
  290. int non_blocking);
  291. int iser_reg_page_vec(struct iser_conn *ib_conn,
  292. struct iser_page_vec *page_vec,
  293. struct iser_mem_reg *mem_reg);
  294. void iser_unreg_mem(struct iser_mem_reg *mem_reg);
  295. int iser_post_recvl(struct iser_conn *ib_conn);
  296. int iser_post_recvm(struct iser_conn *ib_conn, int count);
  297. int iser_post_send(struct iser_conn *ib_conn, struct iser_tx_desc *tx_desc);
  298. int iser_dma_map_task_data(struct iscsi_iser_task *iser_task,
  299. struct iser_data_buf *data,
  300. enum iser_data_dir iser_dir,
  301. enum dma_data_direction dma_dir);
  302. void iser_dma_unmap_task_data(struct iscsi_iser_task *iser_task);
  303. int iser_initialize_task_headers(struct iscsi_task *task,
  304. struct iser_tx_desc *tx_desc);
  305. int iser_alloc_rx_descriptors(struct iser_conn *ib_conn);
  306. #endif