/* * CDDL HEADER START * * The contents of this file are subject to the terms of the * Common Development and Distribution License, Version 1.0 only * (the "License"). You may not use this file except in compliance * with the License. * * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE * or http://www.opensolaris.org/os/licensing. * See the License for the specific language governing permissions * and limitations under the License. * * When distributing Covered Code, include this CDDL HEADER in each * file and include the License file at usr/src/OPENSOLARIS.LICENSE. * If applicable, add the following below this CDDL HEADER, with the * fields enclosed by brackets "[]" replaced with your own identifying * information: Portions Copyright [yyyy] [name of copyright owner] * * CDDL HEADER END */ /* Copyright (c) 1984, 1986, 1987, 1988, 1989 AT&T */ /* All Rights Reserved */ /* * Copyright 2005 Sun Microsystems, Inc. All rights reserved. * Use is subject to license terms. */ #ifndef _SYS_STRSUBR_H #define _SYS_STRSUBR_H #pragma ident "%Z%%M% %I% %E% SMI" /* SVr4.0 1.17 */ /* * WARNING: * Everything in this file is private, belonging to the * STREAMS subsystem. The only guarantee made about the * contents of this file is that if you include it, your * code will not port to the next release. */ #include #include #include #include #include #include #include #include #include #ifdef __cplusplus extern "C" { #endif /* * In general, the STREAMS locks are disjoint; they are only held * locally, and not simultaneously by a thread. However, module * code, including at the stream head, requires some locks to be * acquired in order for its safety. * 1. Stream level claim. This prevents the value of q_next * from changing while module code is executing. * 2. Queue level claim. This prevents the value of q_ptr * from changing while put or service code is executing. * In addition, it provides for queue single-threading * for QPAIR and PERQ MT-safe modules. * 3. Stream head lock. May be held by the stream head module * to implement a read/write/open/close monitor. * Note: that the only types of twisted stream supported are * the pipe and transports which have read and write service * procedures on both sides of the twist. * 4. Queue lock. May be acquired by utility routines on * behalf of a module. */ /* * In general, sd_lock protects the consistency of the stdata * structure. Additionally, it is used with sd_monitor * to implement an open/close monitor. In particular, it protects * the following fields: * sd_iocblk * sd_flag * sd_copyflag * sd_iocid * sd_iocwait * sd_sidp * sd_pgidp * sd_wroff * sd_rerror * sd_werror * sd_pushcnt * sd_sigflags * sd_siglist * sd_pollist * sd_mark * sd_closetime * sd_wakeq * sd_uiordq * sd_uiowrq * sd_maxblk * * The following fields are modified only by the allocator, which * has exclusive access to them at that time: * sd_wrq * sd_strtab * * The following field is protected by the overlying file system * code, guaranteeing single-threading of opens: * sd_vnode * * Stream-level locks should be acquired before any queue-level locks * are acquired. * * The stream head write queue lock(sd_wrq) is used to protect the * fields qn_maxpsz and qn_minpsz because freezestr() which is * necessary for strqset() only gets the queue lock. */ /* * Function types for the parameterized stream head. * The msgfunc_t takes the parameters: * msgfunc(vnode_t *vp, mblk_t *mp, strwakeup_t *wakeups, * strsigset_t *firstmsgsigs, strsigset_t *allmsgsigs, * strpollset_t *pollwakeups); * It returns an optional message to be processed by the stream head. * * The parameters for errfunc_t are: * errfunc(vnode *vp, int ispeek, int *clearerr); * It returns an errno and zero if there was no pending error. */ typedef uint_t strwakeup_t; typedef uint_t strsigset_t; typedef short strpollset_t; typedef uintptr_t callbparams_id_t; typedef mblk_t *(*msgfunc_t)(vnode_t *, mblk_t *, strwakeup_t *, strsigset_t *, strsigset_t *, strpollset_t *); typedef int (*errfunc_t)(vnode_t *, int, int *); /* * Per stream sd_lock in putnext may be replaced by per cpu stream_putlocks * each living in a separate cache line. putnext/canputnext grabs only one of * stream_putlocks while strlock() (called on behalf of insertq()/removeq()) * acquires all stream_putlocks. Normally stream_putlocks are only employed * for highly contended streams that have SQ_CIPUT queues in the critical path * (e.g. NFS/UDP stream). * * stream_putlocks are dynamically assigned to stdata structure through * sd_ciputctrl pointer possibly when a stream is already in use. Since * strlock() uses stream_putlocks only under sd_lock acquiring sd_lock when * assigning stream_putlocks to the stream ensures synchronization with * strlock(). * * For lock ordering purposes stream_putlocks are treated as the extension of * sd_lock and are always grabbed right after grabbing sd_lock and released * right before releasing sd_lock except putnext/canputnext where only one of * stream_putlocks locks is used and where it is the first lock to grab. */ typedef struct ciputctrl_str { union _ciput_un { uchar_t pad[64]; struct _ciput_str { kmutex_t ciput_lck; ushort_t ciput_cnt; } ciput_str; } ciput_un; } ciputctrl_t; #define ciputctrl_lock ciput_un.ciput_str.ciput_lck #define ciputctrl_count ciput_un.ciput_str.ciput_cnt /* * Header for a stream: interface to rest of system. */ typedef struct stdata { struct queue *sd_wrq; /* write queue */ struct msgb *sd_iocblk; /* return block for ioctl */ struct vnode *sd_vnode; /* pointer to associated vnode */ struct streamtab *sd_strtab; /* pointer to streamtab for stream */ uint_t sd_flag; /* state/flags */ uint_t sd_iocid; /* ioctl id */ struct pid *sd_sidp; /* controlling session info */ struct pid *sd_pgidp; /* controlling process group info */ ushort_t sd_unused; /* UNUSED, retained for binary */ /* compatibility */ ushort_t sd_wroff; /* write offset */ int sd_rerror; /* error to return on read ops */ int sd_werror; /* error to return on write ops */ int sd_pushcnt; /* number of pushes done on stream */ int sd_sigflags; /* logical OR of all siglist events */ struct strsig *sd_siglist; /* pid linked list to rcv SIGPOLL sig */ struct pollhead sd_pollist; /* list of all pollers to wake up */ struct msgb *sd_mark; /* "marked" message on read queue */ clock_t sd_closetime; /* time to wait to drain q in close */ kmutex_t sd_lock; /* protect head consistency */ kcondvar_t sd_monitor; /* open/close/push/pop monitor */ kcondvar_t sd_iocmonitor; /* ioctl single-threading */ kcondvar_t sd_refmonitor; /* sd_refcnt monitor */ ssize_t sd_qn_minpsz; /* These two fields are a performance */ ssize_t sd_qn_maxpsz; /* enhancements, cache the values in */ /* the stream head so we don't have */ /* to ask the module below the stream */ /* head to get this information. */ struct stdata *sd_mate; /* pointer to twisted stream mate */ kthread_id_t sd_freezer; /* thread that froze stream */ kmutex_t sd_reflock; /* Protects sd_refcnt */ int sd_refcnt; /* number of claimstr */ uint_t sd_wakeq; /* strwakeq()'s copy of sd_flag */ struct queue *sd_struiordq; /* sync barrier struio() read queue */ struct queue *sd_struiowrq; /* sync barrier struio() write queue */ char sd_struiodnak; /* defer NAK of M_IOCTL by rput() */ struct msgb *sd_struionak; /* pointer M_IOCTL mblk(s) to NAK */ caddr_t sd_t_audit_data; /* For audit purposes only */ ssize_t sd_maxblk; /* maximum message block size */ uint_t sd_rput_opt; /* options/flags for strrput */ uint_t sd_wput_opt; /* options/flags for write/putmsg */ uint_t sd_read_opt; /* options/flags for strread */ msgfunc_t sd_rprotofunc; /* rput M_*PROTO routine */ msgfunc_t sd_rmiscfunc; /* rput routine (non-data/proto) */ errfunc_t sd_rderrfunc; /* read side error callback */ errfunc_t sd_wrerrfunc; /* write side error callback */ /* * support for low contention concurrent putnext. */ ciputctrl_t *sd_ciputctrl; uint_t sd_nciputctrl; int sd_anchor; /* position of anchor in stream */ /* * Service scheduling at the stream head. */ kmutex_t sd_qlock; struct queue *sd_qhead; /* Head of queues to be serviced. */ struct queue *sd_qtail; /* Tail of queues to be serviced. */ void *sd_servid; /* Service ID for bckgrnd schedule */ ushort_t sd_svcflags; /* Servicing flags */ short sd_nqueues; /* Number of queues in the list */ kcondvar_t sd_qcv; /* Waiters for qhead to become empty */ kcondvar_t sd_zcopy_wait; uint_t sd_copyflag; /* copy-related flags */ } stdata_t; /* * stdata servicing flags. */ #define STRS_WILLSERVICE 0x01 #define STRS_SCHEDULED 0x02 #define STREAM_NEEDSERVICE(stp) ((stp)->sd_qhead != NULL) /* * stdata flag field defines */ #define IOCWAIT 0x00000001 /* Someone is doing an ioctl */ #define RSLEEP 0x00000002 /* Someone wants to read/recv msg */ #define WSLEEP 0x00000004 /* Someone wants to write */ #define STRPRI 0x00000008 /* An M_PCPROTO is at stream head */ #define STRHUP 0x00000010 /* Device has vanished */ #define STWOPEN 0x00000020 /* waiting for 1st open */ #define STPLEX 0x00000040 /* stream is being multiplexed */ #define STRISTTY 0x00000080 /* stream is a terminal */ #define STRGETINPROG 0x00000100 /* (k)strgetmsg is running */ #define IOCWAITNE 0x00000200 /* STR_NOERROR ioctl running */ #define STRDERR 0x00000400 /* fatal read error from M_ERROR */ #define STWRERR 0x00000800 /* fatal write error from M_ERROR */ #define STRDERRNONPERSIST 0x00001000 /* nonpersistent read errors */ #define STWRERRNONPERSIST 0x00002000 /* nonpersistent write errors */ #define STRCLOSE 0x00004000 /* wait for a close to complete */ #define SNDMREAD 0x00008000 /* used for read notification */ #define OLDNDELAY 0x00010000 /* use old TTY semantics for */ /* NDELAY reads and writes */ /* 0x00020000 unused */ /* 0x00040000 unused */ #define STRTOSTOP 0x00080000 /* block background writes */ /* 0x00100000 unused */ /* 0x00200000 unused */ #define STRMOUNT 0x00400000 /* stream is mounted */ #define STRNOTATMARK 0x00800000 /* Not at mark (when empty read q) */ #define STRDELIM 0x01000000 /* generate delimited messages */ #define STRATMARK 0x02000000 /* At mark (due to MSGMARKNEXT) */ #define STZCNOTIFY 0x04000000 /* wait for zerocopy mblk to be acked */ #define STRPLUMB 0x08000000 /* push/pop pending */ #define STREOF 0x10000000 /* End-of-file indication */ #define STREOPENFAIL 0x20000000 /* indicates if re-open has failed */ #define STRMATE 0x40000000 /* this stream is a mate */ #define STRHASLINKS 0x80000000 /* I_LINKs under this stream */ /* * Copy-related flags (sd_copyflag), set by SO_COPYOPT. */ #define STZCVMSAFE 0x00000001 /* safe to borrow file (segmapped) */ /* pages instead of bcopy */ #define STZCVMUNSAFE 0x00000002 /* unsafe to borrow file pages */ #define STRCOPYCACHED 0x00000004 /* copy should NOT bypass cache */ /* * Options and flags for strrput (sd_rput_opt) */ #define SR_POLLIN 0x00000001 /* pollwakeup needed for band0 data */ #define SR_SIGALLDATA 0x00000002 /* Send SIGPOLL for all M_DATA */ #define SR_CONSOL_DATA 0x00000004 /* Consolidate M_DATA onto q_last */ #define SR_IGN_ZEROLEN 0x00000008 /* Ignore zero-length M_DATA */ /* * Options and flags for strwrite/strputmsg (sd_wput_opt) */ #define SW_SIGPIPE 0x00000001 /* Send SIGPIPE for write error */ #define SW_RECHECK_ERR 0x00000002 /* Recheck errors in strwrite loop */ #define SW_SNDZERO 0x00000004 /* send 0-length msg down pipe/FIFO */ /* * Options and flags for strread (sd_read_opt) */ #define RD_MSGDIS 0x00000001 /* read msg discard */ #define RD_MSGNODIS 0x00000002 /* read msg no discard */ #define RD_PROTDAT 0x00000004 /* read M_[PC]PROTO contents as data */ #define RD_PROTDIS 0x00000008 /* discard M_[PC]PROTO blocks and */ /* retain data blocks */ /* * Flags parameter for strsetrputhooks() and strsetwputhooks(). * These flags define the interface for setting the above internal * flags in sd_rput_opt and sd_wput_opt. */ #define SH_CONSOL_DATA 0x00000001 /* Consolidate M_DATA onto q_last */ #define SH_SIGALLDATA 0x00000002 /* Send SIGPOLL for all M_DATA */ #define SH_IGN_ZEROLEN 0x00000004 /* Drop zero-length M_DATA */ #define SH_SIGPIPE 0x00000100 /* Send SIGPIPE for write error */ #define SH_RECHECK_ERR 0x00000200 /* Recheck errors in strwrite loop */ /* * Each queue points to a sync queue (the inner perimeter) which keeps * track of the number of threads that are inside a given queue (sq_count) * and also is used to implement the asynchronous putnext * (by queuing messages if the queue can not be entered.) * * Messages are queued on sq_head/sq_tail including deferred qwriter(INNER) * messages. The sq_head/sq_tail list is a singly-linked list with * b_queue recording the queue and b_prev recording the function to * be called (either the put procedure or a qwriter callback function.) * * The sq_count counter tracks the number of threads that are * executing inside the perimeter or (in the case of outer perimeters) * have some work queued for them relating to the perimeter. The sq_rmqcount * counter tracks the subset which are in removeq() (usually invoked from * qprocsoff(9F)). * * In addition a module writer can declare that the module has an outer * perimeter (by setting D_MTOUTPERIM) in which case all inner perimeter * syncq's for the module point (through sq_outer) to an outer perimeter * syncq. The outer perimeter consists of the doubly linked list (sq_onext and * sq_oprev) linking all the inner perimeter syncq's with out outer perimeter * syncq. This is used to implement qwriter(OUTER) (an asynchronous way of * getting exclusive access at the outer perimeter) and outer_enter/exit * which are used by the framework to acquire exclusive access to the outer * perimeter during open and close of modules that have set D_MTOUTPERIM. * * In the inner perimeter case sq_save is available for use by machine * dependent code. sq_head/sq_tail are used to queue deferred messages on * the inner perimeter syncqs and to queue become_writer requests on the * outer perimeter syncqs. * * Note: machine dependent optimized versions of putnext may depend * on the order of sq_flags and sq_count (so that they can e.g. * read these two fields in a single load instruction.) * * Per perimeter SQLOCK/sq_count in putnext/put may be replaced by per cpu * sq_putlocks/sq_putcounts each living in a separate cache line. Obviously * sq_putlock[x] protects sq_putcount[x]. putnext/put routine will grab only 1 * of sq_putlocks and update only 1 of sq_putcounts. strlock() and many * other routines in strsubr.c and ddi.c will grab all sq_putlocks (as well as * SQLOCK) and figure out the count value as the sum of sq_count and all of * sq_putcounts. The idea is to make critical fast path -- putnext -- much * faster at the expense of much less often used slower path like * strlock(). One known case where entersq/strlock is executed pretty often is * SpecWeb but since IP is SQ_CIOC and socket TCP/IP stream is nextless * there's no need to grab multiple sq_putlocks and look at sq_putcounts. See * strsubr.c for more comments. * * Note regular SQLOCK and sq_count are still used in many routines * (e.g. entersq(), rwnext()) in the same way as before sq_putlocks were * introduced. * * To understand when all sq_putlocks need to be held and all sq_putcounts * need to be added up one needs to look closely at putnext code. Basically if * a routine like e.g. wait_syncq() needs to be sure that perimeter is empty * all sq_putlocks/sq_putcounts need to be held/added up. On the other hand * there's no need to hold all sq_putlocks and count all sq_putcounts in * routines like leavesq()/dropsq() and etc. since the are usually exit * counterparts of entersq/outer_enter() and etc. which have already either * prevented put entry poins from executing or did not care about put * entrypoints. entersq() doesn't need to care about sq_putlocks/sq_putcounts * if the entry point has a shared access since put has the highest degree of * concurrency and such entersq() does not intend to block out put * entrypoints. * * Before sq_putcounts were introduced the standard way to wait for perimeter * to become empty was: * * mutex_enter(SQLOCK(sq)); * while (sq->sq_count > 0) { * sq->sq_flags |= SQ_WANTWAKEUP; * cv_wait(&sq->sq_wait, SQLOCK(sq)); * } * mutex_exit(SQLOCK(sq)); * * The new way is: * * mutex_enter(SQLOCK(sq)); * count = sq->sq_count; * SQ_PUTLOCKS_ENTER(sq); * SUM_SQ_PUTCOUNTS(sq, count); * while (count != 0) { * sq->sq_flags |= SQ_WANTWAKEUP; * SQ_PUTLOCKS_EXIT(sq); * cv_wait(&sq->sq_wait, SQLOCK(sq)); * count = sq->sq_count; * SQ_PUTLOCKS_ENTER(sq); * SUM_SQ_PUTCOUNTS(sq, count); * } * SQ_PUTLOCKS_EXIT(sq); * mutex_exit(SQLOCK(sq)); * * Note that SQ_WANTWAKEUP is set before dropping SQ_PUTLOCKS. This makes sure * putnext won't skip a wakeup. * * sq_putlocks are treated as the extension of SQLOCK for lock ordering * purposes and are always grabbed right after grabbing SQLOCK and released * right before releasing SQLOCK. This also allows dynamic creation of * sq_putlocks while holding SQLOCK (by making sq_ciputctrl non null even when * the stream is already in use). Only in putnext one of sq_putlocks * is grabbed instead of SQLOCK. putnext return path remembers what counter it * incremented and decrements the right counter on its way out. */ struct syncq { kmutex_t sq_lock; /* atomic access to syncq */ uint16_t sq_count; /* # threads inside */ uint16_t sq_flags; /* state and some type info */ /* * Distributed syncq scheduling * The list of queue's is handled by sq_head and * sq_tail fields. * * The list of events is handled by the sq_evhead and sq_evtail * fields. */ queue_t *sq_head; /* queue of deferred messages */ queue_t *sq_tail; /* queue of deferred messages */ mblk_t *sq_evhead; /* Event message on the syncq */ mblk_t *sq_evtail; uint_t sq_nqueues; /* # of queues on this sq */ /* * Concurrency and condition variables */ uint16_t sq_type; /* type (concurrency) of syncq */ uint16_t sq_rmqcount; /* # threads inside removeq() */ kcondvar_t sq_wait; /* block on this sync queue */ kcondvar_t sq_exitwait; /* waiting for thread to leave the */ /* inner perimeter */ /* * Handling synchronous callbacks such as qtimeout and qbufcall */ ushort_t sq_callbflags; /* flags for callback synchronization */ callbparams_id_t sq_cancelid; /* id of callback being cancelled */ struct callbparams *sq_callbpend; /* Pending callbacks */ /* * Links forming an outer perimeter from one outer syncq and * a set of inner sync queues. */ struct syncq *sq_outer; /* Pointer to outer perimeter */ struct syncq *sq_onext; /* Linked list of syncq's making */ struct syncq *sq_oprev; /* up the outer perimeter. */ /* * support for low contention concurrent putnext. */ ciputctrl_t *sq_ciputctrl; uint_t sq_nciputctrl; /* * Counter for the number of threads wanting to become exclusive. */ uint_t sq_needexcl; /* * These two fields are used for scheduling a syncq for * background processing. The sq_svcflag is protected by * SQLOCK lock. */ struct syncq *sq_next; /* for syncq scheduling */ void * sq_servid; uint_t sq_servcount; /* # pending background threads */ uint_t sq_svcflags; /* Scheduling flags */ clock_t sq_tstamp; /* Time when was enabled */ /* * Maximum priority of the queues on this syncq. */ pri_t sq_pri; }; typedef struct syncq syncq_t; /* * sync queue scheduling flags (for sq_svcflags). */ #define SQ_SERVICE 0x1 /* being serviced */ #define SQ_BGTHREAD 0x2 /* awaiting service by bg thread */ #define SQ_DISABLED 0x4 /* don't put syncq in service list */ /* * FASTPUT bit in sd_count/putcount. */ #define SQ_FASTPUT 0x8000 #define SQ_FASTMASK 0x7FFF /* * sync queue state flags */ #define SQ_EXCL 0x0001 /* exclusive access to inner */ /* perimeter */ #define SQ_BLOCKED 0x0002 /* qprocsoff */ #define SQ_FROZEN 0x0004 /* freezestr */ #define SQ_WRITER 0x0008 /* qwriter(OUTER) pending or running */ #define SQ_MESSAGES 0x0010 /* messages on syncq */ #define SQ_WANTWAKEUP 0x0020 /* do cv_broadcast on sq_wait */ #define SQ_WANTEXWAKEUP 0x0040 /* do cv_broadcast on sq_exitwait */ #define SQ_EVENTS 0x0080 /* Events pending */ #define SQ_QUEUED (SQ_MESSAGES | SQ_EVENTS) #define SQ_FLAGMASK 0x00FF /* * Test a queue to see if inner perimeter is exclusive. */ #define PERIM_EXCL(q) ((q)->q_syncq->sq_flags & SQ_EXCL) /* * If any of these flags are set it is not possible for a thread to * enter a put or service procedure. Instead it must either block * or put the message on the syncq. */ #define SQ_GOAWAY (SQ_EXCL|SQ_BLOCKED|SQ_FROZEN|SQ_WRITER|\ SQ_QUEUED) /* * If any of these flags are set it not possible to drain the syncq */ #define SQ_STAYAWAY (SQ_BLOCKED|SQ_FROZEN|SQ_WRITER) /* * Flags to trigger syncq tail processing. */ #define SQ_TAIL (SQ_QUEUED|SQ_WANTWAKEUP|SQ_WANTEXWAKEUP) /* * Syncq types (stored in sq_type) * The SQ_TYPES_IN_FLAGS (ciput) are also stored in sq_flags * for performance reasons. Thus these type values have to be in the low * 16 bits and not conflict with the sq_flags values above. * * Notes: * - putnext() and put() assume that the put procedures have the highest * degree of concurrency. Thus if any of the SQ_CI* are set then SQ_CIPUT * has to be set. This restriction can be lifted by adding code to putnext * and put that check that sq_count == 0 like entersq does. * - putnext() and put() does currently not handle !SQ_COPUT * - In order to implement !SQ_COCB outer_enter has to be fixed so that * the callback can be cancelled while cv_waiting in outer_enter. * - If SQ_CISVC needs to be implemented, qprocsoff() needs to wait * for the currently running services to stop (wait for QINSERVICE * to go off). disable_svc called from qprcosoff disables only * services that will be run in future. * * All the SQ_CO flags are set when there is no outer perimeter. */ #define SQ_CIPUT 0x0100 /* Concurrent inner put proc */ #define SQ_CISVC 0x0200 /* Concurrent inner svc proc */ #define SQ_CIOC 0x0400 /* Concurrent inner open/close */ #define SQ_CICB 0x0800 /* Concurrent inner callback */ #define SQ_COPUT 0x1000 /* Concurrent outer put proc */ #define SQ_COSVC 0x2000 /* Concurrent outer svc proc */ #define SQ_COOC 0x4000 /* Concurrent outer open/close */ #define SQ_COCB 0x8000 /* Concurrent outer callback */ /* Types also kept in sq_flags for performance */ #define SQ_TYPES_IN_FLAGS (SQ_CIPUT) #define SQ_CI (SQ_CIPUT|SQ_CISVC|SQ_CIOC|SQ_CICB) #define SQ_CO (SQ_COPUT|SQ_COSVC|SQ_COOC|SQ_COCB) #define SQ_TYPEMASK (SQ_CI|SQ_CO) /* * Flag combinations passed to entersq and leavesq to specify the type * of entry point. */ #define SQ_PUT (SQ_CIPUT|SQ_COPUT) #define SQ_SVC (SQ_CISVC|SQ_COSVC) #define SQ_OPENCLOSE (SQ_CIOC|SQ_COOC) #define SQ_CALLBACK (SQ_CICB|SQ_COCB) /* * Other syncq types which are not copied into flags. */ #define SQ_PERMOD 0x01 /* Syncq is PERMOD */ /* * Asynchronous callback qun*** flag. * The mechanism these flags are used in is one where callbacks enter * the perimeter thanks to framework support. To use this mechanism * the q* and qun* flavors of the callback routines must be used. * e.g. qtimeout and quntimeout. The synchronization provided by the flags * avoids deadlocks between blocking qun* routines and the perimeter * lock. */ #define SQ_CALLB_BYPASSED 0x01 /* bypassed callback fn */ /* * Cancel callback mask. * The mask expands as the number of cancelable callback types grows * Note - separate callback flag because different callbacks have * overlapping id space. */ #define SQ_CALLB_CANCEL_MASK (SQ_CANCEL_TOUT|SQ_CANCEL_BUFCALL) #define SQ_CANCEL_TOUT 0x02 /* cancel timeout request */ #define SQ_CANCEL_BUFCALL 0x04 /* cancel bufcall request */ typedef struct callbparams { syncq_t *cbp_sq; void (*cbp_func)(void *); void *cbp_arg; callbparams_id_t cbp_id; uint_t cbp_flags; struct callbparams *cbp_next; size_t cbp_size; } callbparams_t; typedef struct strbufcall { void (*bc_func)(void *); void *bc_arg; size_t bc_size; bufcall_id_t bc_id; struct strbufcall *bc_next; kthread_id_t bc_executor; } strbufcall_t; /* * Structure of list of processes to be sent SIGPOLL/SIGURG signal * on request. The valid S_* events are defined in stropts.h. */ typedef struct strsig { struct pid *ss_pidp; /* pid/pgrp pointer */ pid_t ss_pid; /* positive pid, negative pgrp */ int ss_events; /* S_* events */ struct strsig *ss_next; } strsig_t; /* * bufcall list */ struct bclist { strbufcall_t *bc_head; strbufcall_t *bc_tail; }; /* * Structure used to track mux links and unlinks. */ struct mux_node { major_t mn_imaj; /* internal major device number */ uint16_t mn_indegree; /* number of incoming edges */ struct mux_node *mn_originp; /* where we came from during search */ struct mux_edge *mn_startp; /* where search left off in mn_outp */ struct mux_edge *mn_outp; /* list of outgoing edges */ uint_t mn_flags; /* see below */ }; /* * Flags for mux_nodes. */ #define VISITED 1 /* * Edge structure - a list of these is hung off the * mux_node to represent the outgoing edges. */ struct mux_edge { struct mux_node *me_nodep; /* edge leads to this node */ struct mux_edge *me_nextp; /* next edge */ int me_muxid; /* id of link */ }; /* * Queue info * * The syncq is included here to reduce memory fragmentation * for kernel memory allocators that only allocate in sizes that are * powers of two. If the kernel memory allocator changes this should * be revisited. */ typedef struct queinfo { struct queue qu_rqueue; /* read queue - must be first */ struct queue qu_wqueue; /* write queue - must be second */ struct syncq qu_syncq; /* syncq - must be third */ } queinfo_t; /* * Multiplexed streams info */ typedef struct linkinfo { struct linkblk li_lblk; /* must be first */ struct file *li_fpdown; /* file pointer for lower stream */ struct linkinfo *li_next; /* next in list */ struct linkinfo *li_prev; /* previous in list */ } linkinfo_t; /* * List of syncq's used by freeezestr/unfreezestr */ typedef struct syncql { struct syncql *sql_next; syncq_t *sql_sq; } syncql_t; typedef struct sqlist { syncql_t *sqlist_head; size_t sqlist_size; /* structure size in bytes */ size_t sqlist_index; /* next free entry in array */ syncql_t sqlist_array[4]; /* 4 or more entries */ } sqlist_t; typedef struct perdm { struct perdm *dm_next; syncq_t *dm_sq; struct streamtab *dm_str; uint_t dm_ref; } perdm_t; #define NEED_DM(dmp, qflag) \ (dmp == NULL && (qflag & (QPERMOD | QMTOUTPERIM))) /* * fmodsw_impl_t is used within the kernel. fmodsw is used by * the modules/drivers. The information is copied from fmodsw * defined in the module/driver into the fmodsw_impl_t structure * during the module/driver initialization. */ typedef struct fmodsw_impl fmodsw_impl_t; struct fmodsw_impl { fmodsw_impl_t *f_next; char f_name[FMNAMESZ + 1]; struct streamtab *f_str; uint32_t f_qflag; uint32_t f_sqtype; perdm_t *f_dmp; uint32_t f_ref; uint32_t f_hits; }; typedef enum { FMODSW_HOLD = 0x00000001, FMODSW_LOAD = 0x00000002 } fmodsw_flags_t; typedef struct cdevsw_impl { struct streamtab *d_str; uint32_t d_qflag; uint32_t d_sqtype; perdm_t *d_dmp; } cdevsw_impl_t; /* * Finding related queues */ #define STREAM(q) ((q)->q_stream) #define SQ(rq) ((syncq_t *)((rq) + 2)) /* * Locking macros */ #define QLOCK(q) (&(q)->q_lock) #define SQLOCK(sq) (&(sq)->sq_lock) #define STREAM_PUTLOCKS_ENTER(stp) { \ ASSERT(MUTEX_HELD(&(stp)->sd_lock)); \ if ((stp)->sd_ciputctrl != NULL) { \ int i; \ int nlocks = (stp)->sd_nciputctrl; \ ciputctrl_t *cip = (stp)->sd_ciputctrl; \ for (i = 0; i <= nlocks; i++) { \ mutex_enter(&cip[i].ciputctrl_lock); \ } \ } \ } #define STREAM_PUTLOCKS_EXIT(stp) { \ ASSERT(MUTEX_HELD(&(stp)->sd_lock)); \ if ((stp)->sd_ciputctrl != NULL) { \ int i; \ int nlocks = (stp)->sd_nciputctrl; \ ciputctrl_t *cip = (stp)->sd_ciputctrl; \ for (i = 0; i <= nlocks; i++) { \ mutex_exit(&cip[i].ciputctrl_lock); \ } \ } \ } #define SQ_PUTLOCKS_ENTER(sq) { \ ASSERT(MUTEX_HELD(SQLOCK(sq))); \ if ((sq)->sq_ciputctrl != NULL) { \ int i; \ int nlocks = (sq)->sq_nciputctrl; \ ciputctrl_t *cip = (sq)->sq_ciputctrl; \ ASSERT((sq)->sq_type & SQ_CIPUT); \ for (i = 0; i <= nlocks; i++) { \ mutex_enter(&cip[i].ciputctrl_lock); \ } \ } \ } #define SQ_PUTLOCKS_EXIT(sq) { \ ASSERT(MUTEX_HELD(SQLOCK(sq))); \ if ((sq)->sq_ciputctrl != NULL) { \ int i; \ int nlocks = (sq)->sq_nciputctrl; \ ciputctrl_t *cip = (sq)->sq_ciputctrl; \ ASSERT((sq)->sq_type & SQ_CIPUT); \ for (i = 0; i <= nlocks; i++) { \ mutex_exit(&cip[i].ciputctrl_lock); \ } \ } \ } #define SQ_PUTCOUNT_SETFAST(sq) { \ ASSERT(MUTEX_HELD(SQLOCK(sq))); \ if ((sq)->sq_ciputctrl != NULL) { \ int i; \ int nlocks = (sq)->sq_nciputctrl; \ ciputctrl_t *cip = (sq)->sq_ciputctrl; \ ASSERT((sq)->sq_type & SQ_CIPUT); \ for (i = 0; i <= nlocks; i++) { \ mutex_enter(&cip[i].ciputctrl_lock); \ cip[i].ciputctrl_count |= SQ_FASTPUT; \ mutex_exit(&cip[i].ciputctrl_lock); \ } \ } \ } #define SQ_PUTCOUNT_CLRFAST(sq) { \ ASSERT(MUTEX_HELD(SQLOCK(sq))); \ if ((sq)->sq_ciputctrl != NULL) { \ int i; \ int nlocks = (sq)->sq_nciputctrl; \ ciputctrl_t *cip = (sq)->sq_ciputctrl; \ ASSERT((sq)->sq_type & SQ_CIPUT); \ for (i = 0; i <= nlocks; i++) { \ mutex_enter(&cip[i].ciputctrl_lock); \ cip[i].ciputctrl_count &= ~SQ_FASTPUT; \ mutex_exit(&cip[i].ciputctrl_lock); \ } \ } \ } #ifdef DEBUG #define SQ_PUTLOCKS_HELD(sq) { \ ASSERT(MUTEX_HELD(SQLOCK(sq))); \ if ((sq)->sq_ciputctrl != NULL) { \ int i; \ int nlocks = (sq)->sq_nciputctrl; \ ciputctrl_t *cip = (sq)->sq_ciputctrl; \ ASSERT((sq)->sq_type & SQ_CIPUT); \ for (i = 0; i <= nlocks; i++) { \ ASSERT(MUTEX_HELD(&cip[i].ciputctrl_lock)); \ } \ } \ } #define SUMCHECK_SQ_PUTCOUNTS(sq, countcheck) { \ if ((sq)->sq_ciputctrl != NULL) { \ int i; \ uint_t count = 0; \ int ncounts = (sq)->sq_nciputctrl; \ ASSERT((sq)->sq_type & SQ_CIPUT); \ for (i = 0; i <= ncounts; i++) { \ count += \ (((sq)->sq_ciputctrl[i].ciputctrl_count) & \ SQ_FASTMASK); \ } \ ASSERT(count == (countcheck)); \ } \ } #define SUMCHECK_CIPUTCTRL_COUNTS(ciput, nciput, countcheck) { \ int i; \ uint_t count = 0; \ ASSERT((ciput) != NULL); \ for (i = 0; i <= (nciput); i++) { \ count += (((ciput)[i].ciputctrl_count) & \ SQ_FASTMASK); \ } \ ASSERT(count == (countcheck)); \ } #else /* DEBUG */ #define SQ_PUTLOCKS_HELD(sq) #define SUMCHECK_SQ_PUTCOUNTS(sq, countcheck) #define SUMCHECK_CIPUTCTRL_COUNTS(sq, nciput, countcheck) #endif /* DEBUG */ #define SUM_SQ_PUTCOUNTS(sq, count) { \ if ((sq)->sq_ciputctrl != NULL) { \ int i; \ int ncounts = (sq)->sq_nciputctrl; \ ciputctrl_t *cip = (sq)->sq_ciputctrl; \ ASSERT((sq)->sq_type & SQ_CIPUT); \ for (i = 0; i <= ncounts; i++) { \ (count) += ((cip[i].ciputctrl_count) & \ SQ_FASTMASK); \ } \ } \ } #define CLAIM_QNEXT_LOCK(stp) mutex_enter(&(stp)->sd_lock) #define RELEASE_QNEXT_LOCK(stp) mutex_exit(&(stp)->sd_lock) /* * syncq message manipulation macros. */ /* * Put a message on the queue syncq. * Assumes QLOCK held. */ #define SQPUT_MP(qp, mp) \ { \ qp->q_syncqmsgs++; \ if (qp->q_sqhead == NULL) { \ qp->q_sqhead = qp->q_sqtail = mp; \ } else { \ qp->q_sqtail->b_next = mp; \ qp->q_sqtail = mp; \ } \ } /* * Miscellaneous parameters and flags. */ /* * Default timeout in milliseconds for ioctls and close */ #define STRTIMOUT 15000 /* * Flag values for stream io */ #define WRITEWAIT 0x1 /* waiting for write event */ #define READWAIT 0x2 /* waiting for read event */ #define NOINTR 0x4 /* error is not to be set for signal */ #define GETWAIT 0x8 /* waiting for getmsg event */ /* * These flags need to be unique for stream io name space * and copy modes name space. These flags allow strwaitq * and strdoioctl to proceed as if signals or errors on the stream * head have not occurred; i.e. they will be detected by some other * means. * STR_NOSIG does not allow signals to interrupt the call * STR_NOERROR does not allow stream head read, write or hup errors to * affect the call. When used with strdoioctl(), if a previous ioctl * is pending and times out, STR_NOERROR will cause strdoioctl() to not * return ETIME. If, however, the requested ioctl times out, ETIME * will be returned (use ic_timout instead) * STR_PEEK is used to inform strwaitq that the reader is peeking at data * and that a non-persistent error should not be cleared. * STR_DELAYERR is used to inform strwaitq that it should not check errors * after being awoken since, in addition to an error, there might also be * data queued on the stream head read queue. */ #define STR_NOSIG 0x10 /* Ignore signals during strdoioctl/strwaitq */ #define STR_NOERROR 0x20 /* Ignore errors during strdoioctl/strwaitq */ #define STR_PEEK 0x40 /* Peeking behavior on non-persistent errors */ #define STR_DELAYERR 0x80 /* Do not check errors on return */ /* * Copy modes for tty and I_STR ioctls */ #define U_TO_K 01 /* User to Kernel */ #define K_TO_K 02 /* Kernel to Kernel */ /* * Mux defines. */ #define LINKNORMAL 0x01 /* normal mux link */ #define LINKPERSIST 0x02 /* persistent mux link */ #define LINKTYPEMASK 0x03 /* bitmask of all link types */ #define LINKCLOSE 0x04 /* unlink from strclose */ /* * Definitions of Streams macros and function interfaces. */ /* * Obsolete queue scheduling macros. They are not used anymore, but still kept * here for 3-d party modules and drivers who might still use them. */ #define setqsched() #define qready() 1 #ifdef _KERNEL #define runqueues() #define queuerun() #endif /* compatibility module for style 2 drivers with DR race condition */ #define DRMODNAME "drcompat" /* * Macros dealing with mux_nodes. */ #define MUX_VISIT(X) ((X)->mn_flags |= VISITED) #define MUX_CLEAR(X) ((X)->mn_flags &= (~VISITED)); \ ((X)->mn_originp = NULL) #define MUX_DIDVISIT(X) ((X)->mn_flags & VISITED) /* * Twisted stream macros */ #define STRMATED(X) ((X)->sd_flag & STRMATE) #define STRLOCKMATES(X) if (&((X)->sd_lock) > &(((X)->sd_mate)->sd_lock)) { \ mutex_enter(&((X)->sd_lock)); \ mutex_enter(&(((X)->sd_mate)->sd_lock)); \ } else { \ mutex_enter(&(((X)->sd_mate)->sd_lock)); \ mutex_enter(&((X)->sd_lock)); \ } #define STRUNLOCKMATES(X) mutex_exit(&((X)->sd_lock)); \ mutex_exit(&(((X)->sd_mate)->sd_lock)) #ifdef _KERNEL extern void strinit(void); extern int strdoioctl(struct stdata *, struct strioctl *, int, int, cred_t *, int *); extern void strsendsig(struct strsig *, int, uchar_t, int); extern void str_sendsig(vnode_t *, int, uchar_t, int); extern void strhup(struct stdata *); extern int qattach(queue_t *, dev_t *, int, cred_t *, fmodsw_impl_t *, boolean_t); extern int qreopen(queue_t *, dev_t *, int, cred_t *); extern void qdetach(queue_t *, int, int, cred_t *, boolean_t); extern void enterq(queue_t *); extern void leaveq(queue_t *); extern int putiocd(mblk_t *, caddr_t, int, cred_t *); extern int getiocd(mblk_t *, caddr_t, int); extern struct linkinfo *alloclink(queue_t *, queue_t *, struct file *); extern void lbfree(struct linkinfo *); extern int linkcycle(stdata_t *, stdata_t *); extern struct linkinfo *findlinks(stdata_t *, int, int); extern queue_t *getendq(queue_t *); extern int mlink(vnode_t *, int, int, cred_t *, int *, int); extern int mlink_file(vnode_t *, int, struct file *, cred_t *, int *, int); extern int munlink(struct stdata *, struct linkinfo *, int, cred_t *, int *); extern int munlinkall(struct stdata *, int, cred_t *, int *); extern void mux_addedge(stdata_t *, stdata_t *, int); extern void mux_rmvedge(stdata_t *, int); extern int devflg_to_qflag(struct streamtab *, uint32_t, uint32_t *, uint32_t *); extern void setq(queue_t *, struct qinit *, struct qinit *, perdm_t *, uint32_t, uint32_t, boolean_t); extern perdm_t *hold_dm(struct streamtab *, uint32_t, uint32_t); extern void rele_dm(perdm_t *); extern int strmakectl(struct strbuf *, int32_t, int32_t, mblk_t **); extern int strmakedata(ssize_t *, struct uio *, stdata_t *, int32_t, mblk_t **); extern int strmakemsg(struct strbuf *, ssize_t *, struct uio *, struct stdata *, int32_t, mblk_t **); extern int strgetmsg(vnode_t *, struct strbuf *, struct strbuf *, uchar_t *, int *, int, rval_t *); extern int strputmsg(vnode_t *, struct strbuf *, struct strbuf *, uchar_t, int flag, int fmode); extern int strstartplumb(struct stdata *, int, int); extern void strendplumb(struct stdata *); extern struct streamtab *fifo_getinfo(void); extern int stropen(struct vnode *, dev_t *, int, cred_t *); extern int strclose(struct vnode *, int, cred_t *); extern int strpoll(register struct stdata *, short, int, short *, struct pollhead **); extern void strclean(struct vnode *); extern void str_cn_clean(); /* XXX hook for consoles signal cleanup */ extern int strwrite(struct vnode *, struct uio *, cred_t *); extern int strread(struct vnode *, struct uio *, cred_t *); extern int strioctl(struct vnode *, int, intptr_t, int, int, cred_t *, int *); extern int strrput(queue_t *, mblk_t *); extern int strrput_nondata(queue_t *, mblk_t *); extern mblk_t *strrput_proto(vnode_t *, mblk_t *, strwakeup_t *, strsigset_t *, strsigset_t *, strpollset_t *); extern mblk_t *strrput_misc(vnode_t *, mblk_t *, strwakeup_t *, strsigset_t *, strsigset_t *, strpollset_t *); extern int getiocseqno(void); extern int strwaitbuf(size_t, int); extern int strwaitq(stdata_t *, int, ssize_t, int, clock_t, int *); extern void strctty(struct stdata *); extern void stralloctty(sess_t *, struct stdata *); extern void strfreectty(struct stdata *); extern struct stdata *shalloc(queue_t *); extern void shfree(struct stdata *s); extern queue_t *allocq(void); extern void freeq(queue_t *); extern qband_t *allocband(void); extern void freeband(qband_t *); extern void freebs_enqueue(mblk_t *, dblk_t *); extern void setqback(queue_t *, unsigned char); extern int strcopyin(void *, void *, size_t, int); extern int strcopyout(void *, void *, size_t, int); extern void strsignal(struct stdata *, int, int32_t); extern clock_t str_cv_wait(kcondvar_t *, kmutex_t *, clock_t, int); extern void runbuffcalls(void); extern void disable_svc(queue_t *); extern void remove_runlist(queue_t *); extern void wait_svc(queue_t *); extern void backenable(queue_t *, uchar_t); extern void set_qend(queue_t *); extern void set_qnexthot(queue_t *); extern int strgeterr(stdata_t *, int32_t, int); extern void qenable_locked(queue_t *); extern mblk_t *getq_noenab(queue_t *); extern void rmvq_noenab(queue_t *, mblk_t *); extern void qbackenable(queue_t *, uchar_t); extern void strblock(queue_t *); extern void strunblock(queue_t *); extern int qclaimed(queue_t *); extern int straccess(struct stdata *, enum jcaccess); extern void entersq(syncq_t *, int); extern void leavesq(syncq_t *, int); extern void claimq(queue_t *); extern void releaseq(queue_t *); extern void claimstr(queue_t *); extern void releasestr(queue_t *); extern void removeq(queue_t *); extern void insertq(struct stdata *, queue_t *); extern void fill_syncq(syncq_t *, queue_t *, mblk_t *, void (*)()); extern void drain_syncq(syncq_t *); extern void qfill_syncq(syncq_t *, queue_t *, mblk_t *); extern void qdrain_syncq(syncq_t *, queue_t *); extern int flush_syncq(syncq_t *, queue_t *); extern void wait_sq_svc(syncq_t *); extern void outer_enter(syncq_t *, uint16_t); extern void outer_exit(syncq_t *); extern void qwriter_inner(queue_t *, mblk_t *, void (*)()); extern void qwriter_outer(queue_t *, mblk_t *, void (*)()); extern callbparams_t *callbparams_alloc(syncq_t *, void (*)(void *), void *, int); extern void callbparams_free(syncq_t *, callbparams_t *); extern void callbparams_free_id(syncq_t *, callbparams_id_t, int32_t); extern void qcallbwrapper(void *); extern mblk_t *desballoc(unsigned char *, size_t, uint_t, frtn_t *); extern mblk_t *esballoca(unsigned char *, size_t, uint_t, frtn_t *); extern mblk_t *desballoca(unsigned char *, size_t, uint_t, frtn_t *); extern int do_sendfp(struct stdata *, struct file *, struct cred *); extern int frozenstr(queue_t *); extern size_t xmsgsize(mblk_t *); extern void putnext_tail(syncq_t *, queue_t *, uint32_t); extern void stream_willservice(stdata_t *); extern void stream_runservice(stdata_t *); extern void strmate(vnode_t *, vnode_t *); extern queue_t *strvp2wq(vnode_t *); extern vnode_t *strq2vp(queue_t *); extern mblk_t *allocb_wait(size_t, uint_t, uint_t, int *); extern mblk_t *allocb_cred(size_t, cred_t *); extern mblk_t *allocb_cred_wait(size_t, uint_t, int *, cred_t *); extern mblk_t *allocb_tmpl(size_t, const mblk_t *); extern void mblk_setcred(mblk_t *, cred_t *); void strpollwakeup(vnode_t *, short); extern int putnextctl_wait(queue_t *, int); int kstrputmsg(struct vnode *, mblk_t *, struct uio *, ssize_t, unsigned char, int, int); int kstrgetmsg(struct vnode *, mblk_t **, struct uio *, unsigned char *, int *, clock_t, rval_t *); extern int kstrwritemp(struct vnode *, mblk_t *, ushort_t); void strsetrerror(vnode_t *, int, int, errfunc_t); void strsetwerror(vnode_t *, int, int, errfunc_t); void strseteof(vnode_t *, int); void strflushrq(vnode_t *, int); void strsetrputhooks(vnode_t *, uint_t, msgfunc_t, msgfunc_t); void strsetwputhooks(vnode_t *, uint_t, clock_t); int strwaitmark(vnode_t *); extern void strsignal_nolock(stdata_t *, int, int32_t); struct multidata_s; struct pdesc_s; extern int hcksum_assoc(mblk_t *, struct multidata_s *, struct pdesc_s *, uint32_t, uint32_t, uint32_t, uint32_t, uint32_t, int); extern void hcksum_retrieve(mblk_t *, struct multidata_s *, struct pdesc_s *, uint32_t *, uint32_t *, uint32_t *, uint32_t *, uint32_t *); extern unsigned int bcksum(uchar_t *, int, unsigned int); extern boolean_t is_vmloaned_mblk(mblk_t *, struct multidata_s *, struct pdesc_s *); extern int fmodsw_register(const char *, struct streamtab *, int); extern int fmodsw_unregister(const char *); extern fmodsw_impl_t *fmodsw_find(const char *, fmodsw_flags_t); extern void fmodsw_rele(fmodsw_impl_t *); extern void freemsgchain(mblk_t *); extern mblk_t *copymsgchain(mblk_t *); /* * shared or externally configured data structures */ extern ssize_t strmsgsz; /* maximum stream message size */ extern ssize_t strctlsz; /* maximum size of ctl message */ extern int nstrpush; /* maximum number of pushes allowed */ /* * Bufcalls related variables. */ extern struct bclist strbcalls; /* List of bufcalls */ extern kmutex_t strbcall_lock; /* Protects the list of bufcalls */ extern kcondvar_t strbcall_cv; /* Signaling when a bufcall is added */ extern kcondvar_t bcall_cv; /* wait of executing bufcall completes */ extern frtn_t frnop; extern struct kmem_cache *ciputctrl_cache; extern int n_ciputctrl; extern int max_n_ciputctrl; extern int min_n_ciputctrl; extern cdevsw_impl_t *devimpl; #endif /* _KERNEL */ /* * Note: Use of these macros are restricted to kernel/unix and * intended for the STREAMS framework. * All modules/drivers should include sys/ddi.h. * * Finding related queues */ #define _OTHERQ(q) ((q)->q_flag&QREADR? (q)+1: (q)-1) #define _WR(q) ((q)->q_flag&QREADR? (q)+1: (q)) #define _RD(q) ((q)->q_flag&QREADR? (q): (q)-1) #define _SAMESTR(q) (!((q)->q_flag & QEND)) /* * These are also declared here for modules/drivers that erroneously * include strsubr.h after ddi.h or fail to include ddi.h at all. */ extern struct queue *OTHERQ(queue_t *); /* stream.h */ extern struct queue *RD(queue_t *); extern struct queue *WR(queue_t *); extern int SAMESTR(queue_t *); #ifdef __cplusplus } #endif #endif /* _SYS_STRSUBR_H */