xref: /illumos-gate/usr/src/uts/common/fs/sockfs/socktpi.h (revision bde334a8dbd66dfa70ce4d7fc9dcad6e1ae45fe4)
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 /*
23  * Copyright (c) 2008, 2010, Oracle and/or its affiliates. All rights reserved.
24  */
25 
26 #ifndef _SOCKFS_SOCKTPI_H
27 #define	_SOCKFS_SOCKTPI_H
28 
29 #ifdef	__cplusplus
30 extern "C" {
31 #endif
32 
33 /*
34  * Internal representation used for addresses.
35  */
36 struct soaddr {
37 	struct sockaddr	*soa_sa;	/* Actual address */
38 	t_uscalar_t	soa_len;	/* Length in bytes for kmem_free */
39 	t_uscalar_t	soa_maxlen;	/* Allocated length */
40 };
41 /* Maximum size address for transports that have ADDR_size == 1 */
42 #define	SOA_DEFSIZE	128
43 
44 struct sonode;
45 
46 /*
47  * TPI Sockets
48  * ======================
49  *
50  * A TPI socket can be created by the TPI socket module, or as a
51  * result of fallback. In either case, the TPI related information is
52  * stored in a sotpi_info_t. Sockets that are TPI based from the
53  * beginning will use a sotpi_sonode_t, but fallback case the
54  * sotpi_info_t will be allocated when needed. However, the so_priv
55  * field in the sonode will always point to the sotpi_info_t, and the
56  * structure should only be accessed via so_priv. Use SOTOTPI().
57  *
58  * A TPI socket always corresponds to a VCHR stream representing the
59  * transport provider (e.g. /dev/tcp). This information is retrieved
60  * from the kernel socket configuration table and accessible via
61  * so_sockparams->sp_sdev_info.  sockfs uses this to perform
62  * VOP_ACCESS checks before allowing an open of the transport
63  * provider.
64  *
65  * AF_UNIX Sockets
66  * -------------------------
67  *
68  * When an AF_UNIX socket is bound to a pathname the sockfs creates a
69  * VSOCK vnode in the underlying file system. However, the vnodeops
70  * etc in this VNODE remain those of the underlying file system.
71  * Sockfs uses the v_stream pointer in the underlying file system
72  * VSOCK node to find the sonode bound to the pathname. The bound
73  * pathname vnode is accessed through sti_ux_vp.
74  *
75  * Out of Band Data Handling
76  * -------------------------
77  *
78  * The counts (sti_oobcnt and sti_oobsigcnt) track the number of
79  * urgent indicates that are (logically) queued on the stream head
80  * read queue. The urgent data is queued on the stream head
81  * as follows.
82  *
83  * In the normal case the SIGURG is not generated until
84  * the T_EXDATA_IND arrives at the stream head. However, transports
85  * that have an early indication that urgent data is pending
86  * (e.g. TCP receiving a "new" urgent pointer value) can send up
87  * an M_PCPROTO/SIGURG message to generate the signal early.
88  *
89  * The mark is indicated by either:
90  *  - a T_EXDATA_IND (with no M_DATA b_cont) with MSGMARK set.
91  *    When this message is consumed by sorecvmsg the socket layer
92  *    sets SS_RCVATMARK until data has been consumed past the mark.
93  *  - a message with MSGMARKNEXT set (indicating that the
94  *    first byte of the next message constitutes the mark). When
95  *    the last byte of the MSGMARKNEXT message is consumed in
96  *    the stream head the stream head sets STRATMARK. This flag
97  *    is cleared when at least one byte is read. (Note that
98  *    the MSGMARKNEXT messages can be of zero length when there
99  *    is no previous data to which the marknext can be attached.)
100  *
101  * While the T_EXDATA_IND method is the common case which is used
102  * with all TPI transports, the MSGMARKNEXT method is needed to
103  * indicate the mark when e.g. the TCP urgent byte has not been
104  * received yet but the TCP urgent pointer has made TCP generate
105  * the M_PCSIG/SIGURG.
106  *
107  * The signal (the M_PCSIG carrying the SIGURG) and the mark
108  * indication can not be delivered as a single message, since
109  * the signal should be delivered as high priority and any mark
110  * indication must flow with the data. This implies that immediately
111  * when the SIGURG has been delivered if the stream head queue is
112  * empty it is impossible to determine if this will be the position
113  * of the mark. This race condition is resolved by using MSGNOTMARKNEXT
114  * messages and the STRNOTATMARK flag in the stream head. The
115  * SIOCATMARK code calls the stream head to wait for either a
116  * non-empty queue or one of the STR*ATMARK flags being set.
117  * This implies that any transport that is sending M_PCSIG(SIGURG)
118  * should send the appropriate MSGNOTMARKNEXT message (which can be
119  * zero length) after sending an M_PCSIG to prevent SIOCATMARK
120  * from sleeping unnecessarily.
121  */
122 
123 #define	SOTPI_INFO_MAGIC	0x12345678
124 
125 /*
126  * Information used by TPI/STREAMS sockets
127  */
128 typedef struct sotpi_info {
129 	/*
130 	 * These fields are initialized once.
131 	 */
132 	uint32_t	sti_magic;	/* always set to SOTPI_INFO_MAGIC */
133 	dev_t		sti_dev;	/* device the sonode represents */
134 
135 	struct sockparams *sti_orig_sp;	/* in case of fallback; the orig sp */
136 
137 	kmutex_t	sti_plumb_lock;	/* serializes plumbs, and the related */
138 					/* so_pushcnt */
139 	short		sti_pushcnt;	/* Number of modules above "sockmod" */
140 
141 	kcondvar_t	sti_ack_cv;	/* wait for TPI acks */
142 
143 	uint8_t
144 		sti_laddr_valid : 1,	/* sti_laddr valid for user */
145 		sti_faddr_valid : 1,	/* sti_faddr valid for user */
146 		sti_faddr_noxlate : 1,	/* No xlation of faddr for AF_UNIX */
147 
148 		sti_direct : 1,		/* transport is directly below */
149 
150 		sti_pad_to_bit7 : 4;
151 
152 	mblk_t	*sti_ack_mp;		/* TPI ack received from below */
153 	mblk_t	*sti_unbind_mp;		/* Preallocated T_UNBIND_REQ message */
154 
155 	time_t  sti_atime;		/* time of last access */
156 	time_t  sti_mtime;		/* time of last modification */
157 	time_t  sti_ctime;		/* time of last attributes change */
158 
159 	ushort_t sti_delayed_error;	/* From T_uderror_ind */
160 	mblk_t	*sti_eaddr_mp;		/* for so_delayed_error */
161 					/* put here for delayed processing  */
162 
163 	mblk_t	*sti_conn_ind_head;	/* b_next list of T_CONN_IND */
164 	mblk_t	*sti_conn_ind_tail;
165 
166 	uint_t	sti_oobsigcnt;		/* Number of SIGURG generated */
167 	uint_t	sti_oobcnt;		/* Number of T_EXDATA_IND queued */
168 
169 	/* From T_info_ack */
170 	t_uscalar_t	sti_tsdu_size;
171 	t_uscalar_t	sti_etsdu_size;
172 	t_scalar_t	sti_addr_size;
173 	t_uscalar_t	sti_opt_size;
174 	t_uscalar_t	sti_tidu_size;
175 	t_scalar_t	sti_serv_type;
176 
177 	/* From T_capability_ack */
178 	t_uscalar_t	sti_acceptor_id;
179 
180 	/* Internal provider information */
181 	struct tpi_provinfo	*sti_provinfo;
182 
183 	/*
184 	 * The local and remote addresses have multiple purposes
185 	 * but one of the key reasons for their existence and careful
186 	 * tracking in sockfs is to support getsockname and getpeername
187 	 * when the transport does not handle the TI_GET*NAME ioctls
188 	 * and caching when it does (signalled by valid bits in so_state).
189 	 * When all transports support the new TPI (with T_ADDR_REQ)
190 	 * we can revisit this code.
191 	 *
192 	 * The other usage of sti_faddr is to keep the "connected to"
193 	 * address for datagram sockets.
194 	 *
195 	 * Finally, for AF_UNIX both local and remote addresses are used
196 	 * to record the sockaddr_un since we use a separate namespace
197 	 * in the loopback transport.
198 	 */
199 	struct soaddr sti_laddr;	/* Local address */
200 	struct soaddr sti_faddr;	/* Peer address */
201 #define	sti_laddr_sa		sti_laddr.soa_sa
202 #define	sti_faddr_sa		sti_faddr.soa_sa
203 #define	sti_laddr_len		sti_laddr.soa_len
204 #define	sti_faddr_len		sti_faddr.soa_len
205 #define	sti_laddr_maxlen	sti_laddr.soa_maxlen
206 #define	sti_faddr_maxlen	sti_faddr.soa_maxlen
207 
208 	/*
209 	 * For AF_UNIX sockets:
210 	 *
211 	 * sti_ux_laddr/faddr records the internal addresses used with the
212 	 * transport. sti_ux_vp and v_stream->sd_vnode form the
213 	 * cross-linkage between the underlying fs vnode corresponding
214 	 * to the bound sockaddr_un and the socket node.
215 	 *
216 	 * sti_ux_taddr holds the result of translations done in
217 	 * so_ux_addr_xlate(), which may or may not be the same as
218 	 * sti_ux_faddr (which is our connected peer address).
219 	 */
220 	struct so_ux_addr sti_ux_laddr; /* laddr bound with the transport */
221 	struct so_ux_addr sti_ux_faddr; /* connected peer address */
222 	struct so_ux_addr sti_ux_taddr; /* temporary address for sendmsg */
223 	struct vnode	*sti_ux_bound_vp; /* bound AF_UNIX file system vnode */
224 	struct sonode	*sti_next_so;	/* next sonode on socklist	*/
225 	struct sonode	*sti_prev_so;	/* previous sonode on socklist	*/
226 	mblk_t	*sti_discon_ind_mp;	/* T_DISCON_IND received from below */
227 
228 	/*
229 	 * For NL7C sockets:
230 	 *
231 	 * sti_nl7c_flags	the NL7C state of URL processing.
232 	 *
233 	 * sti_nl7c_rcv_mp	mblk_t chain of already received data to be
234 	 *			passed up to the app after NL7C gives up on
235 	 *			a socket.
236 	 *
237 	 * sti_nl7c_rcv_rval	returned rval for last mblk_t from above.
238 	 *
239 	 * sti_nl7c_uri		the URI currently being processed.
240 	 *
241 	 * sti_nl7c_rtime	URI request gethrestime_sec().
242 	 *
243 	 * sti_nl7c_addr	pointer returned by nl7c_addr_lookup().
244 	 */
245 	uint64_t	sti_nl7c_flags;
246 	mblk_t		*sti_nl7c_rcv_mp;
247 	int64_t		sti_nl7c_rcv_rval;
248 	void		*sti_nl7c_uri;
249 	time_t		sti_nl7c_rtime;
250 	void		*sti_nl7c_addr;
251 } sotpi_info_t;
252 
253 struct T_capability_ack;
254 
255 extern sonodeops_t sotpi_sonodeops;
256 
257 extern int	socktpi_init(void);
258 extern int	sotpi_convert_sonode(struct sonode *, struct sockparams *,
259 		    boolean_t *, queue_t **, struct cred *);
260 extern void	sotpi_revert_sonode(struct sonode *, struct cred *);
261 extern void	sotpi_update_state(struct sonode *, struct T_capability_ack *,
262 		    struct sockaddr *, socklen_t, struct sockaddr *, socklen_t,
263 		    short);
264 
265 extern sotpi_info_t 	*sotpi_sototpi(struct sonode *);
266 #ifdef DEBUG
267 #define	SOTOTPI(so)	(sotpi_sototpi(so))
268 #else
269 #define	SOTOTPI(so)	((sotpi_info_t *)(so)->so_priv)
270 #endif
271 
272 /* for consumers outside sockfs */
273 #define	_SOTOTPI(so)	((sotpi_info_t *)(so)->so_priv)
274 
275 #ifdef	__cplusplus
276 }
277 #endif
278 
279 #endif /* _SOCKFS_SOCKTPI_H */
280