xref: /titanic_44/usr/src/uts/common/sys/idm/idm_impl.h (revision 4eaa471005973e11a6110b69fe990530b3b95a38)
1 /*
2  * CDDL HEADER START
3  *
4  * The contents of this file are subject to the terms of the
5  * Common Development and Distribution License (the "License").
6  * You may not use this file except in compliance with the License.
7  *
8  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9  * or http://www.opensolaris.org/os/licensing.
10  * See the License for the specific language governing permissions
11  * and limitations under the License.
12  *
13  * When distributing Covered Code, include this CDDL HEADER in each
14  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15  * If applicable, add the following below this CDDL HEADER, with the
16  * fields enclosed by brackets "[]" replaced with your own identifying
17  * information: Portions Copyright [yyyy] [name of copyright owner]
18  *
19  * CDDL HEADER END
20  */
21 /*
22  * Copyright 2009 Sun Microsystems, Inc.  All rights reserved.
23  * Use is subject to license terms.
24  */
25 #ifndef	_IDM_IMPL_H_
26 #define	_IDM_IMPL_H_
27 
28 #ifdef	__cplusplus
29 extern "C" {
30 #endif
31 
32 #include <sys/avl.h>
33 #include <sys/socket_impl.h>
34 
35 /*
36  * IDM lock order:
37  *
38  * idm_taskid_table_lock, idm_task_t.idt_mutex
39  */
40 
41 #define	CF_LOGIN_READY		0x00000001
42 #define	CF_INITIAL_LOGIN	0x00000002
43 #define	CF_ERROR		0x80000000
44 
45 typedef enum {
46 	CONN_TYPE_INI = 1,
47 	CONN_TYPE_TGT
48 } idm_conn_type_t;
49 
50 /*
51  * Watchdog interval in seconds
52  */
53 #define	IDM_WD_INTERVAL			5
54 
55 /*
56  * Timeout period before the client "keepalive" callback is invoked in
57  * seconds if the connection is idle.
58  */
59 #define	IDM_TRANSPORT_KEEPALIVE_IDLE_TIMEOUT	20
60 
61 /*
62  * Timeout period before a TRANSPORT_FAIL event is generated in seconds
63  * if the connection is idle.
64  */
65 #define	IDM_TRANSPORT_FAIL_IDLE_TIMEOUT	30
66 
67 /*
68  * IDM reference count structure.  Audit code is shamelessly adapted
69  * from CIFS server.
70  */
71 
72 #define	REFCNT_AUDIT_STACK_DEPTH	16
73 #define	REFCNT_AUDIT_BUF_MAX_REC	16
74 
75 typedef struct {
76 	uint32_t		anr_refcnt;
77 	int			anr_depth;
78 	pc_t			anr_stack[REFCNT_AUDIT_STACK_DEPTH];
79 } refcnt_audit_record_t;
80 
81 typedef struct {
82 	int			anb_index;
83 	int			anb_max_index;
84 	refcnt_audit_record_t	anb_records[REFCNT_AUDIT_BUF_MAX_REC];
85 } refcnt_audit_buf_t;
86 
87 #define	REFCNT_AUDIT(_rf_) {				\
88 	refcnt_audit_record_t	*anr;			\
89 							\
90 	anr = (_rf_)->ir_audit_buf.anb_records;		\
91 	anr += (_rf_)->ir_audit_buf.anb_index;		\
92 	(_rf_)->ir_audit_buf.anb_index++;		\
93 	(_rf_)->ir_audit_buf.anb_index &=		\
94 	    (_rf_)->ir_audit_buf.anb_max_index;		\
95 	anr->anr_refcnt = (_rf_)->ir_refcnt;		\
96 	anr->anr_depth = getpcstack(anr->anr_stack,	\
97 	    REFCNT_AUDIT_STACK_DEPTH);			\
98 }
99 
100 struct idm_refcnt_s;
101 
102 typedef void (idm_refcnt_cb_t)(void *ref_obj);
103 
104 typedef enum {
105 	REF_NOWAIT,
106 	REF_WAIT_SYNC,
107 	REF_WAIT_ASYNC
108 } idm_refcnt_wait_t;
109 
110 typedef struct idm_refcnt_s {
111 	int			ir_refcnt;
112 	void			*ir_referenced_obj;
113 	idm_refcnt_wait_t	ir_waiting;
114 	kmutex_t		ir_mutex;
115 	kcondvar_t		ir_cv;
116 	idm_refcnt_cb_t		*ir_cb;
117 	refcnt_audit_buf_t	ir_audit_buf;
118 } idm_refcnt_t;
119 
120 /*
121  * connection parameters - These parameters would be populated at
122  * connection create, or during key-value negotiation at login
123  */
124 typedef struct idm_conn_params_s {
125 	uint32_t		max_recv_dataseglen;
126 	uint32_t		max_xmit_dataseglen;
127 	uint32_t		conn_login_max;
128 	uint32_t		conn_login_interval;
129 	boolean_t		nonblock_socket;
130 } idm_conn_param_t;
131 
132 typedef struct idm_svc_s {
133 	list_node_t		is_list_node;
134 	kmutex_t		is_mutex;
135 	kcondvar_t		is_cv;
136 	kmutex_t		is_count_mutex;
137 	kcondvar_t		is_count_cv;
138 	idm_refcnt_t		is_refcnt;
139 	int			is_online;
140 	/* transport-specific service components */
141 	void			*is_so_svc;
142 	void			*is_iser_svc;
143 	idm_svc_req_t		is_svc_req;
144 } idm_svc_t;
145 
146 #define	ISCSI_MAX_TSIH_LEN	6	/* 0x%04x */
147 #define	ISCSI_MAX_ISID_LEN	ISCSI_ISID_LEN * 2
148 
149 typedef struct idm_conn_s {
150 	list_node_t		ic_list_node;
151 	void			*ic_handle;
152 	idm_refcnt_t		ic_refcnt;
153 	idm_svc_t		*ic_svc_binding; /* Target conn. only */
154 	idm_sockaddr_t 		ic_ini_dst_addr;
155 	struct sockaddr_storage	ic_laddr;	/* conn local address */
156 	struct sockaddr_storage	ic_raddr;	/* conn remote address */
157 
158 	/*
159 	 * the target_name, initiator_name, initiator session
160 	 * identifier and target session identifying handle
161 	 * are only used for target connections.
162 	 */
163 	char			ic_target_name[ISCSI_MAX_NAME_LEN + 1];
164 	char			ic_initiator_name[ISCSI_MAX_NAME_LEN + 1];
165 	char			ic_tsih[ISCSI_MAX_TSIH_LEN + 1];
166 	char			ic_isid[ISCSI_MAX_ISID_LEN + 1];
167 	idm_conn_state_t	ic_state;
168 	idm_conn_state_t	ic_last_state;
169 	sm_audit_buf_t		ic_state_audit;
170 	kmutex_t		ic_state_mutex;
171 	kcondvar_t		ic_state_cv;
172 	uint32_t		ic_state_flags;
173 	timeout_id_t		ic_state_timeout;
174 	struct idm_conn_s	*ic_reinstate_conn; /* For conn reinst. */
175 	struct idm_conn_s	*ic_logout_conn; /* For other conn logout */
176 	taskq_t			*ic_state_taskq;
177 	int			ic_pdu_events;
178 	boolean_t		ic_login_info_valid;
179 	boolean_t		ic_rdma_extensions;
180 	uint16_t		ic_login_cid;
181 
182 	kmutex_t		ic_mutex;
183 	kcondvar_t		ic_cv;
184 	idm_status_t		ic_conn_sm_status;
185 
186 	boolean_t		ic_ffp;
187 	boolean_t		ic_keepalive;
188 	uint32_t		ic_internal_cid;
189 
190 	uint32_t		ic_conn_flags;
191 	idm_conn_type_t		ic_conn_type;
192 	idm_conn_ops_t		ic_conn_ops;
193 	idm_transport_ops_t	*ic_transport_ops;
194 	idm_transport_type_t	ic_transport_type;
195 	int			ic_transport_hdrlen;
196 	void			*ic_transport_private;
197 	idm_conn_param_t	ic_conn_params;
198 	/*
199 	 * Save client callback to interpose idm callback
200 	 */
201 	idm_pdu_cb_t		*ic_client_callback;
202 	clock_t			ic_timestamp;
203 } idm_conn_t;
204 
205 #define	IDM_CONN_HEADER_DIGEST	0x00000001
206 #define	IDM_CONN_DATA_DIGEST	0x00000002
207 #define	IDM_CONN_USE_SCOREBOARD	0x00000004
208 
209 #define	IDM_CONN_ISINI(ICI_IC)	((ICI_IC)->ic_conn_type == CONN_TYPE_INI)
210 #define	IDM_CONN_ISTGT(ICI_IC)	((ICI_IC)->ic_conn_type == CONN_TYPE_TGT)
211 
212 /*
213  * An IDM target task can transfer data using multiple buffers. The task
214  * will maintain a list of buffers, and each buffer will contain the relative
215  * offset of the transfer and a pointer to the next buffer in the list.
216  *
217  * Note on client private data:
218  * idt_private is intended to be a pointer to some sort of client-
219  * specific state.
220  *
221  * idt_client_handle is a more generic client-private piece of data that can
222  * be used by the client for the express purpose of task lookup.  The driving
223  * use case for this is for the client to store the initiator task tag for
224  * a given task so that it may be more easily retrieved for task management.
225  *
226  * The key take away here is that clients should never call
227  * idm_task_find_by_handle in the performance path.
228  *
229  * An initiator will require only one buffer per task, the offset will be 0.
230  */
231 
232 typedef struct idm_task_s {
233 	idm_conn_t		*idt_ic;	/* Associated connection */
234 	/* connection type is in idt_ic->ic_conn_type */
235 	kmutex_t		idt_mutex;
236 	void			*idt_private;	/* Client private data */
237 	uintptr_t		idt_client_handle;	/* Client private */
238 	uint32_t		idt_tt;		/* Task tag */
239 	uint32_t		idt_r2t_ttt;	/* R2T Target Task tag */
240 	idm_task_state_t	idt_state;
241 	idm_refcnt_t		idt_refcnt;
242 
243 	/*
244 	 * Statistics
245 	 */
246 	int			idt_tx_to_ini_start;
247 	int			idt_tx_to_ini_done;
248 	int			idt_rx_from_ini_start;
249 	int			idt_rx_from_ini_done;
250 	int			idt_tx_bytes;	/* IDM_CONN_USE_SCOREBOARD */
251 	int			idt_rx_bytes;	/* IDM_CONN_USE_SCOREBOARD */
252 
253 	uint32_t		idt_exp_datasn;	/* expected datasn */
254 	uint32_t		idt_exp_rttsn;	/* expected rttsn */
255 	list_t			idt_inbufv;	/* chunks of IN buffers */
256 	list_t			idt_outbufv;	/* chunks of OUT buffers */
257 
258 	/*
259 	 * Transport header, which describes this tasks remote tagged buffer
260 	 */
261 	int			idt_transport_hdrlen;
262 	void			*idt_transport_hdr;
263 } idm_task_t;
264 
265 int idm_task_constructor(void *task_void, void *arg, int flags);
266 void idm_task_destructor(void *task_void, void *arg);
267 
268 #define	IDM_TASKIDS_MAX		16384
269 #define	IDM_BUF_MAGIC		0x49425546	/* "IBUF" */
270 
271 /* Protect with task mutex */
272 typedef struct idm_buf_s {
273 	uint32_t	idb_magic;	/* "IBUF" */
274 
275 	/*
276 	 * Note: idm_tx_link *must* be the second element in the list for
277 	 * proper TX PDU ordering.
278 	 */
279 	list_node_t	idm_tx_link;	/* link in a list of TX objects */
280 
281 	list_node_t	idb_buflink;	/* link in a multi-buffer data xfer */
282 	idm_conn_t	*idb_ic;	/* Associated connection */
283 	void		*idb_buf;	/* data */
284 	uint64_t	idb_buflen;	/* length of buffer */
285 	size_t		idb_bufoffset;	/* offset in a multi-buffer xfer */
286 	boolean_t	idb_bufalloc;  /* true if alloc'd in idm_buf_alloc */
287 	/*
288 	 * DataPDUInOrder=Yes, so to track that the PDUs in a sequence are sent
289 	 * in continuously increasing address order, check that offsets for a
290 	 * single buffer xfer are in order.
291 	 */
292 	uint32_t	idb_exp_offset;
293 	size_t		idb_xfer_len;	/* Current requested xfer len */
294 	void		*idb_buf_private; /* transport-specific buf handle */
295 	void		*idb_reg_private; /* transport-specific reg handle */
296 	void		*idb_bufptr; /* transport-specific bcopy pointer */
297 	boolean_t	idb_bufbcopy;	/* true if bcopy required */
298 
299 	idm_buf_cb_t	*idb_buf_cb;	/* Data Completion Notify, tgt only */
300 	void		*idb_cb_arg;	/* Client private data */
301 	idm_task_t	*idb_task_binding;
302 	timespec_t	idb_xfer_start;
303 	timespec_t	idb_xfer_done;
304 	boolean_t	idb_in_transport;
305 	boolean_t	idb_tx_thread;		/* Sockets only */
306 	iscsi_hdr_t	idb_data_hdr_tmpl;	/* Sockets only */
307 	idm_status_t	idb_status;
308 } idm_buf_t;
309 
310 typedef enum {
311 	BP_CHECK_QUICK,
312 	BP_CHECK_THOROUGH,
313 	BP_CHECK_ASSERT
314 } idm_bufpat_check_type_t;
315 
316 #define	BUFPAT_MATCH(bc_bufpat, bc_idb) 		\
317 	((bufpat->bufpat_idb == bc_idb) &&		\
318 	    (bufpat->bufpat_bufmagic == IDM_BUF_MAGIC))
319 
320 typedef struct idm_bufpat_s {
321 	void		*bufpat_idb;
322 	uint32_t	bufpat_bufmagic;
323 	uint32_t	bufpat_offset;
324 } idm_bufpat_t;
325 
326 #define	PDU_MAX_IOVLEN	12
327 #define	IDM_PDU_MAGIC	0x49504455	/* "IPDU" */
328 
329 typedef struct idm_pdu_s {
330 	uint32_t	isp_magic;	/* "IPDU" */
331 
332 	/*
333 	 * Internal - Order is vital.  idm_tx_link *must* be the second
334 	 * element in this structure for proper TX PDU ordering.
335 	 */
336 	list_node_t	idm_tx_link;
337 
338 	list_node_t	isp_client_lnd;
339 
340 	idm_conn_t	*isp_ic;	/* Must be set */
341 	iscsi_hdr_t	*isp_hdr;
342 	uint_t		isp_hdrlen;
343 	uint8_t		*isp_data;
344 	uint_t		isp_datalen;
345 
346 	/* Transport header */
347 	void		*isp_transport_hdr;
348 	uint32_t	isp_transport_hdrlen;
349 	void		*isp_transport_private;
350 
351 	/*
352 	 * isp_data is used for sending SCSI status, NOP, text, scsi and
353 	 * non-scsi data. Data is received using isp_iov and isp_iovlen
354 	 * to support data over multiple buffers.
355 	 */
356 	void		*isp_private;
357 	idm_pdu_cb_t	*isp_callback;
358 	idm_status_t	isp_status;
359 
360 	/*
361 	 * The following four elements are only used in
362 	 * idm_sorecv_scsidata() currently.
363 	 */
364 	struct iovec	isp_iov[PDU_MAX_IOVLEN];
365 	int		isp_iovlen;
366 	idm_buf_t	*isp_sorx_buf;
367 
368 	/* Implementation data for idm_pdu_alloc and sorx PDU cache */
369 	uint32_t	isp_flags;
370 	uint_t		isp_hdrbuflen;
371 	uint_t		isp_databuflen;
372 } idm_pdu_t;
373 
374 /*
375  * This "generic" object is used when removing an item from the ic_tx_list
376  * in order to determine whether it's an idm_pdu_t or an idm_buf_t
377  */
378 
379 typedef struct {
380 	uint32_t	idm_tx_obj_magic;
381 	/*
382 	 * idm_tx_link *must* be the second element in this structure.
383 	 */
384 	list_node_t	idm_tx_link;
385 } idm_tx_obj_t;
386 
387 
388 #define	IDM_PDU_OPCODE(PDU) \
389 	((PDU)->isp_hdr->opcode & ISCSI_OPCODE_MASK)
390 
391 #define	IDM_PDU_ALLOC		0x00000001
392 #define	IDM_PDU_ADDL_HDR	0x00000002
393 #define	IDM_PDU_ADDL_DATA	0x00000004
394 #define	IDM_PDU_LOGIN_TX	0x00000008
395 
396 #define	OSD_EXT_CDB_AHSLEN	(200 - 15)
397 #define	BIDI_AHS_LENGTH		5
398 #define	IDM_SORX_CACHE_AHSLEN \
399 	(((OSD_EXT_CDB_AHSLEN + 3) + \
400 	    (BIDI_AHS_LENGTH + 3)) / sizeof (uint32_t))
401 #define	IDM_SORX_CACHE_HDRLEN	(sizeof (iscsi_hdr_t) + IDM_SORX_CACHE_AHSLEN)
402 
403 /*
404  * ID pool
405  */
406 
407 #define	IDM_IDPOOL_MAGIC	0x4944504C	/* IDPL */
408 #define	IDM_IDPOOL_MIN_SIZE	64	/* Number of IDs to begin with */
409 #define	IDM_IDPOOL_MAX_SIZE	64 * 1024
410 
411 typedef struct idm_idpool {
412 	uint32_t	id_magic;
413 	kmutex_t	id_mutex;
414 	uint8_t		*id_pool;
415 	uint32_t	id_size;
416 	uint8_t		id_bit;
417 	uint8_t		id_bit_idx;
418 	uint32_t	id_idx;
419 	uint32_t	id_idx_msk;
420 	uint32_t	id_free_counter;
421 	uint32_t	id_max_free_counter;
422 } idm_idpool_t;
423 
424 /*
425  * Global IDM state structure
426  */
427 typedef struct {
428 	kmutex_t	idm_global_mutex;
429 	taskq_t		*idm_global_taskq;
430 	kthread_t	*idm_wd_thread;
431 	kt_did_t	idm_wd_thread_did;
432 	boolean_t	idm_wd_thread_running;
433 	kcondvar_t	idm_wd_cv;
434 	list_t		idm_tgt_svc_list;
435 	kcondvar_t	idm_tgt_svc_cv;
436 	list_t		idm_tgt_conn_list;
437 	int		idm_tgt_conn_count;
438 	list_t		idm_ini_conn_list;
439 	kmem_cache_t	*idm_buf_cache;
440 	kmem_cache_t	*idm_task_cache;
441 	krwlock_t	idm_taskid_table_lock;
442 	idm_task_t	**idm_taskid_table;
443 	uint32_t	idm_taskid_next;
444 	uint32_t	idm_taskid_max;
445 	idm_idpool_t	idm_conn_id_pool;
446 	kmem_cache_t	*idm_sotx_pdu_cache;
447 	kmem_cache_t	*idm_sorx_pdu_cache;
448 	kmem_cache_t	*idm_so_128k_buf_cache;
449 } idm_global_t;
450 
451 idm_global_t	idm; /* Global state */
452 
453 int
454 idm_idpool_create(idm_idpool_t	*pool);
455 
456 void
457 idm_idpool_destroy(idm_idpool_t *pool);
458 
459 int
460 idm_idpool_alloc(idm_idpool_t *pool, uint16_t *id);
461 
462 void
463 idm_idpool_free(idm_idpool_t *pool, uint16_t id);
464 
465 void
466 idm_pdu_rx(idm_conn_t *ic, idm_pdu_t *pdu);
467 
468 void
469 idm_pdu_tx_forward(idm_conn_t *ic, idm_pdu_t *pdu);
470 
471 boolean_t
472 idm_pdu_rx_forward_ffp(idm_conn_t *ic, idm_pdu_t *pdu);
473 
474 void
475 idm_pdu_rx_forward(idm_conn_t *ic, idm_pdu_t *pdu);
476 
477 void
478 idm_pdu_tx_protocol_error(idm_conn_t *ic, idm_pdu_t *pdu);
479 
480 void
481 idm_pdu_rx_protocol_error(idm_conn_t *ic, idm_pdu_t *pdu);
482 
483 void idm_parse_login_rsp(idm_conn_t *ic, idm_pdu_t *logout_req_pdu,
484     boolean_t rx);
485 
486 void idm_parse_logout_req(idm_conn_t *ic, idm_pdu_t *logout_req_pdu,
487     boolean_t rx);
488 
489 void idm_parse_logout_rsp(idm_conn_t *ic, idm_pdu_t *login_rsp_pdu,
490     boolean_t rx);
491 
492 idm_status_t idm_svc_conn_create(idm_svc_t *is, idm_transport_type_t type,
493     idm_conn_t **ic_result);
494 
495 void idm_svc_conn_destroy(idm_conn_t *ic);
496 
497 idm_status_t idm_ini_conn_finish(idm_conn_t *ic);
498 
499 idm_status_t idm_tgt_conn_finish(idm_conn_t *ic);
500 
501 idm_conn_t *idm_conn_create_common(idm_conn_type_t conn_type,
502     idm_transport_type_t tt, idm_conn_ops_t *conn_ops);
503 
504 void idm_conn_destroy_common(idm_conn_t *ic);
505 
506 void idm_conn_close(idm_conn_t *ic);
507 
508 uint32_t idm_cid_alloc(void);
509 
510 void idm_cid_free(uint32_t cid);
511 
512 uint32_t idm_crc32c(void *address, unsigned long length);
513 
514 uint32_t idm_crc32c_continued(void *address, unsigned long length,
515     uint32_t crc);
516 
517 void idm_listbuf_insert(list_t *lst, idm_buf_t *buf);
518 
519 int idm_task_compare(const void *v1, const void *v2);
520 
521 idm_conn_t *idm_lookup_conn(uint8_t *isid, uint16_t tsih, uint16_t cid);
522 
523 #ifdef	__cplusplus
524 }
525 #endif
526 
527 #endif /* _IDM_IMPL_H_ */
528