xref: /titanic_50/usr/src/uts/common/sys/idm/idm_impl.h (revision 6f786ace10b9c0c7c5515e525fb660fbccfda6a3)
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_dataseglen;
126 } idm_conn_param_t;
127 
128 typedef struct idm_svc_s {
129 	list_node_t		is_list_node;
130 	kmutex_t		is_mutex;
131 	kcondvar_t		is_cv;
132 	kmutex_t		is_count_mutex;
133 	kcondvar_t		is_count_cv;
134 	idm_refcnt_t		is_refcnt;
135 	int			is_online;
136 	/* transport-specific service components */
137 	void			*is_so_svc;
138 	void			*is_iser_svc;
139 	idm_svc_req_t		is_svc_req;
140 } idm_svc_t;
141 
142 typedef struct idm_conn_s {
143 	list_node_t		ic_list_node;
144 	void			*ic_handle;
145 	idm_refcnt_t		ic_refcnt;
146 	idm_svc_t		*ic_svc_binding; /* Target conn. only */
147 	idm_sockaddr_t 		ic_ini_dst_addr;
148 	struct sockaddr_storage	ic_laddr;	/* conn local address */
149 	struct sockaddr_storage	ic_raddr;	/* conn remote address */
150 	idm_conn_state_t	ic_state;
151 	idm_conn_state_t	ic_last_state;
152 	sm_audit_buf_t		ic_state_audit;
153 	kmutex_t		ic_state_mutex;
154 	kcondvar_t		ic_state_cv;
155 	uint32_t		ic_state_flags;
156 	timeout_id_t		ic_state_timeout;
157 	struct idm_conn_s	*ic_reinstate_conn; /* For conn reinst. */
158 	struct idm_conn_s	*ic_logout_conn; /* For other conn logout */
159 	taskq_t			*ic_state_taskq;
160 	int			ic_pdu_events;
161 	boolean_t		ic_login_info_valid;
162 	boolean_t		ic_rdma_extensions;
163 	uint16_t		ic_login_cid;
164 
165 	kmutex_t		ic_mutex;
166 	kcondvar_t		ic_cv;
167 	idm_status_t		ic_conn_sm_status;
168 
169 	boolean_t		ic_ffp;
170 	boolean_t		ic_keepalive;
171 	uint32_t		ic_internal_cid;
172 
173 	uint32_t		ic_conn_flags;
174 	idm_conn_type_t		ic_conn_type;
175 	idm_conn_ops_t		ic_conn_ops;
176 	idm_transport_ops_t	*ic_transport_ops;
177 	idm_transport_type_t	ic_transport_type;
178 	int			ic_transport_hdrlen;
179 	void			*ic_transport_private;
180 	idm_conn_param_t	ic_conn_params;
181 	/*
182 	 * Save client callback to interpose idm callback
183 	 */
184 	idm_pdu_cb_t		*ic_client_callback;
185 	clock_t			ic_timestamp;
186 } idm_conn_t;
187 
188 #define	IDM_CONN_HEADER_DIGEST	0x00000001
189 #define	IDM_CONN_DATA_DIGEST	0x00000002
190 #define	IDM_CONN_USE_SCOREBOARD	0x00000004
191 
192 #define	IDM_CONN_ISINI(ICI_IC)	((ICI_IC)->ic_conn_type == CONN_TYPE_INI)
193 #define	IDM_CONN_ISTGT(ICI_IC)	((ICI_IC)->ic_conn_type == CONN_TYPE_TGT)
194 
195 /*
196  * An IDM target task can transfer data using multiple buffers. The task
197  * will maintain a list of buffers, and each buffer will contain the relative
198  * offset of the transfer and a pointer to the next buffer in the list.
199  *
200  * Note on client private data:
201  * idt_private is intended to be a pointer to some sort of client-
202  * specific state.
203  *
204  * idt_client_handle is a more generic client-private piece of data that can
205  * be used by the client for the express purpose of task lookup.  The driving
206  * use case for this is for the client to store the initiator task tag for
207  * a given task so that it may be more easily retrieved for task management.
208  *
209  * The key take away here is that clients should never call
210  * idm_task_find_by_handle in the performance path.
211  *
212  * An initiator will require only one buffer per task, the offset will be 0.
213  */
214 
215 typedef struct idm_task_s {
216 	idm_conn_t		*idt_ic;	/* Associated connection */
217 	/* connection type is in idt_ic->ic_conn_type */
218 	kmutex_t		idt_mutex;
219 	void			*idt_private;	/* Client private data */
220 	uintptr_t		idt_client_handle;	/* Client private */
221 	uint32_t		idt_tt;		/* Task tag */
222 	uint32_t		idt_r2t_ttt;	/* R2T Target Task tag */
223 	idm_task_state_t	idt_state;
224 	idm_refcnt_t		idt_refcnt;
225 
226 	/*
227 	 * Statistics
228 	 */
229 	int			idt_tx_to_ini_start;
230 	int			idt_tx_to_ini_done;
231 	int			idt_rx_from_ini_start;
232 	int			idt_rx_from_ini_done;
233 	int			idt_tx_bytes;	/* IDM_CONN_USE_SCOREBOARD */
234 	int			idt_rx_bytes;	/* IDM_CONN_USE_SCOREBOARD */
235 
236 	uint32_t		idt_exp_datasn;	/* expected datasn */
237 	uint32_t		idt_exp_rttsn;	/* expected rttsn */
238 	list_t			idt_inbufv;	/* chunks of IN buffers */
239 	list_t			idt_outbufv;	/* chunks of OUT buffers */
240 
241 	/*
242 	 * Transport header, which describes this tasks remote tagged buffer
243 	 */
244 	int			idt_transport_hdrlen;
245 	void			*idt_transport_hdr;
246 } idm_task_t;
247 
248 int idm_task_constructor(void *task_void, void *arg, int flags);
249 void idm_task_destructor(void *task_void, void *arg);
250 
251 #define	IDM_TASKIDS_MAX		16384
252 #define	IDM_BUF_MAGIC		0x49425546	/* "IBUF" */
253 
254 /* Protect with task mutex */
255 typedef struct idm_buf_s {
256 	uint32_t	idb_magic;	/* "IBUF" */
257 
258 	/*
259 	 * Note: idm_tx_link *must* be the second element in the list for
260 	 * proper TX PDU ordering.
261 	 */
262 	list_node_t	idm_tx_link;	/* link in a list of TX objects */
263 
264 	list_node_t	idb_buflink;	/* link in a multi-buffer data xfer */
265 	idm_conn_t	*idb_ic;	/* Associated connection */
266 	void		*idb_buf;	/* data */
267 	uint64_t	idb_buflen;	/* length of buffer */
268 	size_t		idb_bufoffset;	/* offset in a multi-buffer xfer */
269 	boolean_t	idb_bufalloc;  /* true if alloc'd in idm_buf_alloc */
270 	/*
271 	 * DataPDUInOrder=Yes, so to track that the PDUs in a sequence are sent
272 	 * in continuously increasing address order, check that offsets for a
273 	 * single buffer xfer are in order.
274 	 */
275 	uint32_t	idb_exp_offset;
276 	size_t		idb_xfer_len;	/* Current requested xfer len */
277 	void		*idb_buf_private; /* transport-specific buf handle */
278 	void		*idb_reg_private; /* transport-specific reg handle */
279 	void		*idb_bufptr; /* transport-specific bcopy pointer */
280 	boolean_t	idb_bufbcopy;	/* true if bcopy required */
281 
282 	idm_buf_cb_t	*idb_buf_cb;	/* Data Completion Notify, tgt only */
283 	void		*idb_cb_arg;	/* Client private data */
284 	idm_task_t	*idb_task_binding;
285 	timespec_t	idb_xfer_start;
286 	timespec_t	idb_xfer_done;
287 	boolean_t	idb_in_transport;
288 	boolean_t	idb_tx_thread;		/* Sockets only */
289 	iscsi_hdr_t	idb_data_hdr_tmpl;	/* Sockets only */
290 	idm_status_t	idb_status;
291 } idm_buf_t;
292 
293 typedef enum {
294 	BP_CHECK_QUICK,
295 	BP_CHECK_THOROUGH,
296 	BP_CHECK_ASSERT
297 } idm_bufpat_check_type_t;
298 
299 #define	BUFPAT_MATCH(bc_bufpat, bc_idb) 		\
300 	((bufpat->bufpat_idb == bc_idb) &&		\
301 	    (bufpat->bufpat_bufmagic == IDM_BUF_MAGIC))
302 
303 typedef struct idm_bufpat_s {
304 	void		*bufpat_idb;
305 	uint32_t	bufpat_bufmagic;
306 	uint32_t	bufpat_offset;
307 } idm_bufpat_t;
308 
309 #define	PDU_MAX_IOVLEN	12
310 #define	IDM_PDU_MAGIC	0x49504455	/* "IPDU" */
311 
312 typedef struct idm_pdu_s {
313 	uint32_t	isp_magic;	/* "IPDU" */
314 
315 	/*
316 	 * Internal - Order is vital.  idm_tx_link *must* be the second
317 	 * element in this structure for proper TX PDU ordering.
318 	 */
319 	list_node_t	idm_tx_link;
320 
321 	list_node_t	isp_client_lnd;
322 
323 	idm_conn_t	*isp_ic;	/* Must be set */
324 	iscsi_hdr_t	*isp_hdr;
325 	uint_t		isp_hdrlen;
326 	uint8_t		*isp_data;
327 	uint_t		isp_datalen;
328 
329 	/* Transport header */
330 	void		*isp_transport_hdr;
331 	uint32_t	isp_transport_hdrlen;
332 	void		*isp_transport_private;
333 
334 	/*
335 	 * isp_data is used for sending SCSI status, NOP, text, scsi and
336 	 * non-scsi data. Data is received using isp_iov and isp_iovlen
337 	 * to support data over multiple buffers.
338 	 */
339 	void		*isp_private;
340 	idm_pdu_cb_t	*isp_callback;
341 	idm_status_t	isp_status;
342 
343 	/*
344 	 * The following four elements are only used in
345 	 * idm_sorecv_scsidata() currently.
346 	 */
347 	struct iovec	isp_iov[PDU_MAX_IOVLEN];
348 	int		isp_iovlen;
349 	idm_buf_t	*isp_sorx_buf;
350 
351 	/* Implementation data for idm_pdu_alloc and sorx PDU cache */
352 	uint32_t	isp_flags;
353 	uint_t		isp_hdrbuflen;
354 	uint_t		isp_databuflen;
355 } idm_pdu_t;
356 
357 /*
358  * This "generic" object is used when removing an item from the ic_tx_list
359  * in order to determine whether it's an idm_pdu_t or an idm_buf_t
360  */
361 
362 typedef struct {
363 	uint32_t	idm_tx_obj_magic;
364 	/*
365 	 * idm_tx_link *must* be the second element in this structure.
366 	 */
367 	list_node_t	idm_tx_link;
368 } idm_tx_obj_t;
369 
370 
371 #define	IDM_PDU_OPCODE(PDU) \
372 	((PDU)->isp_hdr->opcode & ISCSI_OPCODE_MASK)
373 
374 #define	IDM_PDU_ALLOC		0x00000001
375 #define	IDM_PDU_ADDL_HDR	0x00000002
376 #define	IDM_PDU_ADDL_DATA	0x00000004
377 #define	IDM_PDU_LOGIN_TX	0x00000008
378 
379 #define	OSD_EXT_CDB_AHSLEN	(200 - 15)
380 #define	BIDI_AHS_LENGTH		5
381 #define	IDM_SORX_CACHE_AHSLEN \
382 	(((OSD_EXT_CDB_AHSLEN + 3) + \
383 	    (BIDI_AHS_LENGTH + 3)) / sizeof (uint32_t))
384 #define	IDM_SORX_CACHE_HDRLEN	(sizeof (iscsi_hdr_t) + IDM_SORX_CACHE_AHSLEN)
385 
386 /*
387  * ID pool
388  */
389 
390 #define	IDM_IDPOOL_MAGIC	0x4944504C	/* IDPL */
391 #define	IDM_IDPOOL_MIN_SIZE	64	/* Number of IDs to begin with */
392 #define	IDM_IDPOOL_MAX_SIZE	64 * 1024
393 
394 typedef struct idm_idpool {
395 	uint32_t	id_magic;
396 	kmutex_t	id_mutex;
397 	uint8_t		*id_pool;
398 	uint32_t	id_size;
399 	uint8_t		id_bit;
400 	uint8_t		id_bit_idx;
401 	uint32_t	id_idx;
402 	uint32_t	id_idx_msk;
403 	uint32_t	id_free_counter;
404 	uint32_t	id_max_free_counter;
405 } idm_idpool_t;
406 
407 /*
408  * Global IDM state structure
409  */
410 typedef struct {
411 	kmutex_t	idm_global_mutex;
412 	taskq_t		*idm_global_taskq;
413 	kthread_t	*idm_wd_thread;
414 	kt_did_t	idm_wd_thread_did;
415 	boolean_t	idm_wd_thread_running;
416 	kcondvar_t	idm_wd_cv;
417 	list_t		idm_tgt_svc_list;
418 	kcondvar_t	idm_tgt_svc_cv;
419 	list_t		idm_tgt_conn_list;
420 	int		idm_tgt_conn_count;
421 	list_t		idm_ini_conn_list;
422 	kmem_cache_t	*idm_buf_cache;
423 	kmem_cache_t	*idm_task_cache;
424 	krwlock_t	idm_taskid_table_lock;
425 	idm_task_t	**idm_taskid_table;
426 	uint32_t	idm_taskid_next;
427 	uint32_t	idm_taskid_max;
428 	idm_idpool_t	idm_conn_id_pool;
429 	kmem_cache_t	*idm_sotx_pdu_cache;
430 	kmem_cache_t	*idm_sorx_pdu_cache;
431 	kmem_cache_t	*idm_so_128k_buf_cache;
432 } idm_global_t;
433 
434 idm_global_t	idm; /* Global state */
435 
436 int
437 idm_idpool_create(idm_idpool_t	*pool);
438 
439 void
440 idm_idpool_destroy(idm_idpool_t *pool);
441 
442 int
443 idm_idpool_alloc(idm_idpool_t *pool, uint16_t *id);
444 
445 void
446 idm_idpool_free(idm_idpool_t *pool, uint16_t id);
447 
448 void
449 idm_pdu_rx(idm_conn_t *ic, idm_pdu_t *pdu);
450 
451 void
452 idm_pdu_tx_forward(idm_conn_t *ic, idm_pdu_t *pdu);
453 
454 boolean_t
455 idm_pdu_rx_forward_ffp(idm_conn_t *ic, idm_pdu_t *pdu);
456 
457 void
458 idm_pdu_rx_forward(idm_conn_t *ic, idm_pdu_t *pdu);
459 
460 void
461 idm_pdu_tx_protocol_error(idm_conn_t *ic, idm_pdu_t *pdu);
462 
463 void
464 idm_pdu_rx_protocol_error(idm_conn_t *ic, idm_pdu_t *pdu);
465 
466 void idm_parse_login_rsp(idm_conn_t *ic, idm_pdu_t *logout_req_pdu,
467     boolean_t rx);
468 
469 void idm_parse_logout_req(idm_conn_t *ic, idm_pdu_t *logout_req_pdu,
470     boolean_t rx);
471 
472 void idm_parse_logout_rsp(idm_conn_t *ic, idm_pdu_t *login_rsp_pdu,
473     boolean_t rx);
474 
475 idm_status_t idm_svc_conn_create(idm_svc_t *is, idm_transport_type_t type,
476     idm_conn_t **ic_result);
477 
478 void idm_svc_conn_destroy(idm_conn_t *ic);
479 
480 idm_status_t idm_ini_conn_finish(idm_conn_t *ic);
481 
482 idm_status_t idm_tgt_conn_finish(idm_conn_t *ic);
483 
484 idm_conn_t *idm_conn_create_common(idm_conn_type_t conn_type,
485     idm_transport_type_t tt, idm_conn_ops_t *conn_ops);
486 
487 void idm_conn_destroy_common(idm_conn_t *ic);
488 
489 void idm_conn_close(idm_conn_t *ic);
490 
491 uint32_t idm_cid_alloc(void);
492 
493 void idm_cid_free(uint32_t cid);
494 
495 uint32_t idm_crc32c(void *address, unsigned long length);
496 
497 uint32_t idm_crc32c_continued(void *address, unsigned long length,
498     uint32_t crc);
499 
500 void idm_listbuf_insert(list_t *lst, idm_buf_t *buf);
501 
502 int idm_task_compare(const void *v1, const void *v2);
503 
504 idm_conn_t *idm_lookup_conn(uint8_t *isid, uint16_t tsih, uint16_t cid);
505 
506 #ifdef	__cplusplus
507 }
508 #endif
509 
510 #endif /* _IDM_IMPL_H_ */
511