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 2007 Sun Microsystems, Inc. All rights reserved.
24 * Use is subject to license terms.
25 */
26
27 #include <sys/types.h>
28 #include <sys/systm.h>
29 #include <sys/cred.h>
30 #include <sys/modctl.h>
31 #include <sys/vfs.h>
32 #include <sys/vfs_opreg.h>
33 #include <sys/sysmacros.h>
34 #include <sys/cmn_err.h>
35 #include <sys/stat.h>
36 #include <sys/errno.h>
37 #include <sys/kmem.h>
38 #include <sys/file.h>
39 #include <sys/kstat.h>
40 #include <sys/port_impl.h>
41 #include <sys/task.h>
42 #include <sys/project.h>
43
44 /*
45 * Event Ports can be shared across threads or across processes.
46 * Every thread/process can use an own event port or a group of them
47 * can use a single port. A major request was also to get the ability
48 * to submit user-defined events to a port. The idea of the
49 * user-defined events is to use the event ports for communication between
50 * threads/processes (like message queues). User defined-events are queued
51 * in a port with the same priority as other event types.
52 *
53 * Events are delivered only once. The thread/process which is waiting
54 * for events with the "highest priority" (priority here is related to the
55 * internal strategy to wakeup waiting threads) will retrieve the event,
56 * all other threads/processes will not be notified. There is also
57 * the requirement to have events which should be submitted immediately
58 * to all "waiting" threads. That is the main task of the alert event.
59 * The alert event is submitted by the application to a port. The port
60 * changes from a standard mode to the alert mode. Now all waiting threads
61 * will be awaken immediately and they will return with the alert event.
62 * Threads trying to retrieve events from a port in alert mode will
63 * return immediately with the alert event.
64 *
65 *
66 * An event port is like a kernel queue, which accept events submitted from
67 * user level as well as events submitted from kernel sub-systems. Sub-systems
68 * able to submit events to a port are the so-called "event sources".
69 * Current event sources:
70 * PORT_SOURCE_AIO : events submitted per transaction completion from
71 * POSIX-I/O framework.
72 * PORT_SOURCE_TIMER : events submitted when a timer fires
73 * (see timer_create(3RT)).
74 * PORT_SOURCE_FD : events submitted per file descriptor (see poll(2)).
75 * PORT_SOURCE_ALERT : events submitted from user. This is not really a
76 * single event, this is actually a port mode
77 * (see port_alert(3c)).
78 * PORT_SOURCE_USER : events submitted by applications with
79 * port_send(3c) or port_sendn(3c).
80 * PORT_SOURCE_FILE : events submitted per file being watched for file
81 * change events (see port_create(3c).
82 *
83 * There is a user API implemented in the libc library as well as a
84 * kernel API implemented in port_subr.c in genunix.
85 * The available user API functions are:
86 * port_create() : create a port as a file descriptor of portfs file system
87 * The standard close(2) function closes a port.
88 * port_associate() : associate a file descriptor with a port to be able to
89 * retrieve events from that file descriptor.
90 * port_dissociate(): remove the association of a file descriptor with a port.
91 * port_alert() : set/unset a port in alert mode
92 * port_send() : send an event of type PORT_SOURCE_USER to a port
93 * port_sendn() : send an event of type PORT_SOURCE_USER to a list of ports
94 * port_get() : retrieve a single event from a port
95 * port_getn() : retrieve a list of events from a port
96 *
97 * The available kernel API functions are:
98 * port_allocate_event(): allocate an event slot/structure of/from a port
99 * port_init_event() : set event data in the event structure
100 * port_send_event() : send event to a port
101 * port_free_event() : deliver allocated slot/structure back to a port
102 * port_associate_ksource(): associate a kernel event source with a port
103 * port_dissociate_ksource(): dissociate a kernel event source from a port
104 *
105 * The libc implementation consists of small functions which pass the
106 * arguments to the kernel using the "portfs" system call. It means, all the
107 * synchronisation work is being done in the kernel. The "portfs" system
108 * call loads the portfs file system into the kernel.
109 *
110 * PORT CREATION
111 * The first function to be used is port_create() which internally creates
112 * a vnode and a portfs node. The portfs node is represented by the port_t
113 * structure, which again includes all the data necessary to control a port.
114 * port_create() returns a file descriptor, which needs to be used in almost
115 * all other event port functions.
116 * The maximum number of ports per system is controlled by the resource
117 * control: project:port-max-ids.
118 *
119 * EVENT GENERATION
120 * The second step is the triggering of events, which could be sent to a port.
121 * Every event source implements an own method to generate events for a port:
122 * PORT_SOURCE_AIO:
123 * The sigevent structure of the standard POSIX-IO functions
124 * was extended by an additional notification type.
125 * Standard notification types:
126 * SIGEV_NONE, SIGEV_SIGNAL and SIGEV_THREAD
127 * Event ports introduced now SIGEV_PORT.
128 * The notification type SIGEV_PORT specifies that a structure
129 * of type port_notify_t has to be attached to the sigev_value.
130 * The port_notify_t structure contains the event port file
131 * descriptor and a user-defined pointer.
132 * Internally the AIO implementation will use the kernel API
133 * functions to allocate an event port slot per transaction (aiocb)
134 * and sent the event to the port as soon as the transaction completes.
135 * All the events submitted per transaction are of type
136 * PORT_SOURCE_AIO.
137 * PORT_SOURCE_TIMER:
138 * The timer_create() function uses the same method as the
139 * PORT_SOURCE_AIO event source. It also uses the sigevent structure
140 * to deliver the port information.
141 * Internally the timer code will allocate a single event slot/struct
142 * per timer and it will send the timer event as soon as the timer
143 * fires. If the timer-fired event is not delivered to the application
144 * before the next period elapsed, then an overrun counter will be
145 * incremented. The timer event source uses a callback function to
146 * detect the delivery of the event to the application. At that time
147 * the timer callback function will update the event overrun counter.
148 * PORT_SOURCE_FD:
149 * This event source uses the port_associate() function to allocate
150 * an event slot/struct from a port. The application defines in the
151 * events argument of port_associate() the type of events which it is
152 * interested on.
153 * The internal pollwakeup() function is used by all the file
154 * systems --which are supporting the VOP_POLL() interface- to notify
155 * the upper layer (poll(2), devpoll(7d) and now event ports) about
156 * the event triggered (see valid events in poll(2)).
157 * The pollwakeup() function forwards the event to the layer registered
158 * to receive the current event.
159 * The port_dissociate() function can be used to free the allocated
160 * event slot from the port. Anyway, file descriptors deliver events
161 * only one time and remain deactivated until the application
162 * reactivates the association of a file descriptor with port_associate().
163 * If an associated file descriptor is closed then the file descriptor
164 * will be dissociated automatically from the port.
165 *
166 * PORT_SOURCE_ALERT:
167 * This event type is generated when the port was previously set in
168 * alert mode using the port_alert() function.
169 * A single alert event is delivered to every thread which tries to
170 * retrieve events from a port.
171 * PORT_SOURCE_USER:
172 * This type of event is generated from user level using the port_send()
173 * function to send a user event to a port or the port_sendn() function
174 * to send an event to a list of ports.
175 * PORT_SOURCE_FILE:
176 * This event source uses the port_associate() interface to register
177 * a file to be monitored for changes. The file name that needs to be
178 * monitored is specified in the file_obj_t structure, a pointer to which
179 * is passed as an argument. The event types to be monitored are specified
180 * in the events argument.
181 * A file events monitor is represented internal per port per object
182 * address(the file_obj_t pointer). Which means there can be multiple
183 * watches registered on the same file using different file_obj_t
184 * structure pointer. With the help of the FEM(File Event Monitoring)
185 * hooks, the file's vnode ops are intercepted and relevant events
186 * delivered. The port_dissociate() function is used to de-register a
187 * file events monitor on a file. When the specified file is
188 * removed/renamed, the file events watch/monitor is automatically
189 * removed.
190 *
191 * EVENT DELIVERY / RETRIEVING EVENTS
192 * Events remain in the port queue until:
193 * - the application uses port_get() or port_getn() to retrieve events,
194 * - the event source cancel the event,
195 * - the event port is closed or
196 * - the process exits.
197 * The maximal number of events in a port queue is the maximal number
198 * of event slots/structures which can be allocated by event sources.
199 * The allocation of event slots/structures is controlled by the resource
200 * control: process.port-max-events.
201 * The port_get() function retrieves a single event and the port_getn()
202 * function retrieves a list of events.
203 * Events are classified as shareable and non-shareable events across processes.
204 * Non-shareable events are invisible for the port_get(n)() functions of
205 * processes other than the owner of the event.
206 * Shareable event types are:
207 * PORT_SOURCE_USER events
208 * This type of event is unconditionally shareable and without
209 * limitations. If the parent process sends a user event and closes
210 * the port afterwards, the event remains in the port and the child
211 * process will still be able to retrieve the user event.
212 * PORT_SOURCE_ALERT events
213 * This type of event is shareable between processes.
214 * Limitation: The alert mode of the port is removed if the owner
215 * (process which set the port in alert mode) of the
216 * alert event closes the port.
217 * PORT_SOURCE_FD events
218 * This type of event is conditional shareable between processes.
219 * After fork(2) all forked file descriptors are shareable between
220 * the processes. The child process is allowed to retrieve events
221 * from the associated file descriptors and it can also re-associate
222 * the fd with the port.
223 * Limitations: The child process is not allowed to dissociate
224 * the file descriptor from the port. Only the
225 * owner (process) of the association is allowed to
226 * dissociate the file descriptor from the port.
227 * If the owner of the association closes the port
228 * the association will be removed.
229 * PORT_SOURCE_AIO events
230 * This type of event is not shareable between processes.
231 * PORT_SOURCE_TIMER events
232 * This type of event is not shareable between processes.
233 * PORT_SOURCE_FILE events
234 * This type of event is not shareable between processes.
235 *
236 * FORK BEHAVIOUR
237 * On fork(2) the child process inherits all opened file descriptors from
238 * the parent process. This is also valid for port file descriptors.
239 * Associated file descriptors with a port maintain the association across the
240 * fork(2). It means, the child process gets full access to the port and
241 * it can retrieve events from all common associated file descriptors.
242 * Events of file descriptors created and associated with a port after the
243 * fork(2) are non-shareable and can only be retrieved by the same process.
244 *
245 * If the parent or the child process closes an exported port (using fork(2)
246 * or I_SENDFD) all the file descriptors associated with the port by the
247 * process will be dissociated from the port. Events of dissociated file
248 * descriptors as well as all non-shareable events will be discarded.
249 * The other process can continue working with the port as usual.
250 *
251 * CLOSING A PORT
252 * close(2) has to be used to close a port. See FORK BEHAVIOUR for details.
253 *
254 * PORT EVENT STRUCTURES
255 * The global control structure of the event ports framework is port_control_t.
256 * port_control_t keeps track of the number of created ports in the system.
257 * The cache of the port event structures is also located in port_control_t.
258 *
259 * On port_create() the vnode and the portfs node is also created.
260 * The portfs node is represented by the port_t structure.
261 * The port_t structure manages all port specific tasks:
262 * - management of resource control values
263 * - port VOP_POLL interface
264 * - creation time
265 * - uid and gid of the port
266 *
267 * The port_t structure contains the port_queue_t structure.
268 * The port_queue_t structure contains all the data necessary for the
269 * queue management:
270 * - locking
271 * - condition variables
272 * - event counters
273 * - submitted events (represented by port_kevent_t structures)
274 * - threads waiting for event delivery (check portget_t structure)
275 * - PORT_SOURCE_FD cache (managed by the port_fdcache_t structure)
276 * - event source management (managed by the port_source_t structure)
277 * - alert mode management (check port_alert_t structure)
278 *
279 * EVENT MANAGEMENT
280 * The event port file system creates a kmem_cache for internal allocation of
281 * event port structures.
282 *
283 * 1. Event source association with a port:
284 * The first step to do for event sources is to get associated with a port
285 * using the port_associate_ksource() function or adding an entry to the
286 * port_ksource_tab[]. An event source can get dissociated from a port
287 * using the port_dissociate_ksource() function. An entry in the
288 * port_ksource_tab[] implies that the source will be associated
289 * automatically with every new created port.
290 * The event source can deliver a callback function, which is used by the
291 * port to notify the event source about close(2). The idea is that
292 * in such a case the event source should free all allocated resources
293 * and it must return to the port all allocated slots/structures.
294 * The port_close() function will wait until all allocated event
295 * structures/slots are returned to the port.
296 * The callback function is not necessary when the event source does not
297 * maintain local resources, a second condition is that the event source
298 * can guarantee that allocated event slots will be returned without
299 * delay to the port (it will not block and sleep somewhere).
300 *
301 * 2. Reservation of an event slot / event structure
302 * The event port reliability is based on the reservation of an event "slot"
303 * (allocation of an event structure) by the event source as part of the
304 * application call. If the maximal number of event slots is exhausted then
305 * the event source can return a corresponding error code to the application.
306 *
307 * The port_alloc_event() function has to be used by event sources to
308 * allocate an event slot (reserve an event structure). The port_alloc_event()
309 * doesn not block and it will return a 0 value on success or an error code
310 * if it fails.
311 * An argument of port_alloc_event() is a flag which determines the behavior
312 * of the event after it was delivered to the application:
313 * PORT_ALLOC_DEFAULT : event slot becomes free after delivery to the
314 * application.
315 * PORT_ALLOC_PRIVATE : event slot remains under the control of the event
316 * source. This kind of slots can not be used for
317 * event delivery and should only be used internally
318 * by the event source.
319 * PORT_KEV_CACHED : event slot remains under the control of an event
320 * port cache. It does not become free after delivery
321 * to the application.
322 * PORT_ALLOC_SCACHED : event slot remains under the control of the event
323 * source. The event source takes the control over
324 * the slot after the event is delivered to the
325 * application.
326 *
327 * 3. Delivery of events to the event port
328 * Earlier allocated event structure/slot has to be used to deliver
329 * event data to the port. Event source has to use the function
330 * port_send_event(). The single argument is a pointer to the previously
331 * reserved event structure/slot.
332 * The portkev_events field of the port_kevent_t structure can be updated/set
333 * in two ways:
334 * 1. using the port_set_event() function, or
335 * 2. updating the portkev_events field out of the callback function:
336 * The event source can deliver a callback function to the port as an
337 * argument of port_init_event().
338 * One of the arguments of the callback function is a pointer to the
339 * events field, which will be delivered to the application.
340 * (see Delivery of events to the application).
341 * Event structures/slots can be delivered to the event port only one time,
342 * they remain blocked until the data is delivered to the application and the
343 * slot becomes free or it is delivered back to the event source
344 * (PORT_ALLOC_SCACHED). The activation of the callback function mentioned above
345 * is at the same time the indicator for the event source that the event
346 * structure/slot is free for reuse.
347 *
348 * 4. Delivery of events to the application
349 * The events structures/slots delivered by event sources remain in the
350 * port queue until they are retrieved by the application or the port
351 * is closed (exit(2) also closes all opened file descriptors)..
352 * The application uses port_get() or port_getn() to retrieve events from
353 * a port. port_get() retrieves a single event structure/slot and port_getn()
354 * retrieves a list of event structures/slots.
355 * Both functions are able to poll for events and return immediately or they
356 * can specify a timeout value.
357 * Before the events are delivered to the application they are moved to a
358 * second temporary internal queue. The idea is to avoid lock collisions or
359 * contentions of the global queue lock.
360 * The global queue lock is used every time when an event source delivers
361 * new events to the port.
362 * The port_get() and port_getn() functions
363 * a) retrieve single events from the temporary queue,
364 * b) prepare the data to be passed to the application memory,
365 * c) activate the callback function of the event sources:
366 * - to get the latest event data,
367 * - the event source can free all allocated resources associated with the
368 * current event,
369 * - the event source can re-use the current event slot/structure
370 * - the event source can deny the delivery of the event to the application
371 * (e.g. because of the wrong process).
372 * d) put the event back to the temporary queue if the event delivery was denied
373 * e) repeat a) until d) as long as there are events in the queue and
374 * there is enough user space available.
375 *
376 * The loop described above could block for a very long time the global mutex,
377 * to avoid that a second mutex was introduced to synchronized concurrent
378 * threads accessing the temporary queue.
379 */
380
381 static int64_t portfs(int, uintptr_t, uintptr_t, uintptr_t, uintptr_t,
382 uintptr_t);
383
384 static struct sysent port_sysent = {
385 6,
386 SE_ARGC | SE_64RVAL | SE_NOUNLOAD,
387 (int (*)())portfs,
388 };
389
390 static struct modlsys modlsys = {
391 &mod_syscallops, "event ports", &port_sysent
392 };
393
394 #ifdef _SYSCALL32_IMPL
395
396 static int64_t
397 portfs32(uint32_t arg1, int32_t arg2, uint32_t arg3, uint32_t arg4,
398 uint32_t arg5, uint32_t arg6);
399
400 static struct sysent port_sysent32 = {
401 6,
402 SE_ARGC | SE_64RVAL | SE_NOUNLOAD,
403 (int (*)())portfs32,
404 };
405
406 static struct modlsys modlsys32 = {
407 &mod_syscallops32,
408 "32-bit event ports syscalls",
409 &port_sysent32
410 };
411 #endif /* _SYSCALL32_IMPL */
412
413 static struct modlinkage modlinkage = {
414 MODREV_1,
415 { &modlsys,
416 #ifdef _SYSCALL32_IMPL
417 &modlsys32,
418 #endif
419 NULL
420 }
421 };
422
423 port_kstat_t port_kstat = {
424 { "ports", KSTAT_DATA_UINT32 }
425 };
426
427 dev_t portdev;
428 struct vnodeops *port_vnodeops;
429 struct vfs port_vfs;
430
431 extern rctl_hndl_t rc_process_portev;
432 extern rctl_hndl_t rc_project_portids;
433 extern void aio_close_port(void *, int, pid_t, int);
434
435 /*
436 * This table contains a list of event sources which need a static
437 * association with a port (every port).
438 * The last NULL entry in the table is required to detect "end of table".
439 */
440 struct port_ksource port_ksource_tab[] = {
441 {PORT_SOURCE_AIO, aio_close_port, NULL, NULL},
442 {0, NULL, NULL, NULL}
443 };
444
445 /* local functions */
446 static int port_getn(port_t *, port_event_t *, uint_t, uint_t *,
447 port_gettimer_t *);
448 static int port_sendn(int [], int [], uint_t, int, void *, uint_t *);
449 static int port_alert(port_t *, int, int, void *);
450 static int port_dispatch_event(port_t *, int, int, int, uintptr_t, void *);
451 static int port_send(port_t *, int, int, void *);
452 static int port_create(int *);
453 static int port_get_alert(port_alert_t *, port_event_t *);
454 static int port_copy_event(port_event_t *, port_kevent_t *, list_t *);
455 static int *port_errorn(int *, int, int, int);
456 static int port_noshare(void *, int *, pid_t, int, void *);
457 static int port_get_timeout(timespec_t *, timespec_t *, timespec_t **, int *,
458 int);
459 static void port_init(port_t *);
460 static void port_remove_alert(port_queue_t *);
461 static void port_add_ksource_local(port_t *, port_ksource_t *);
462 static void port_check_return_cond(port_queue_t *);
463 static void port_dequeue_thread(port_queue_t *, portget_t *);
464 static portget_t *port_queue_thread(port_queue_t *, uint_t);
465 static void port_kstat_init(void);
466
467 #ifdef _SYSCALL32_IMPL
468 static int port_copy_event32(port_event32_t *, port_kevent_t *, list_t *);
469 #endif
470
471 int
472 _init(void)
473 {
474 static const fs_operation_def_t port_vfsops_template[] = {
475 { NULL, { NULL } }
476 };
477 extern const fs_operation_def_t port_vnodeops_template[];
478 vfsops_t *port_vfsops;
479 int error;
480 major_t major;
481
482 if ((major = getudev()) == (major_t)-1)
483 return (ENXIO);
484 portdev = makedevice(major, 0);
485
486 /* Create a dummy vfs */
487 error = vfs_makefsops(port_vfsops_template, &port_vfsops);
488 if (error) {
489 cmn_err(CE_WARN, "port init: bad vfs ops");
490 return (error);
491 }
492 vfs_setops(&port_vfs, port_vfsops);
493 port_vfs.vfs_flag = VFS_RDONLY;
494 port_vfs.vfs_dev = portdev;
495 vfs_make_fsid(&(port_vfs.vfs_fsid), portdev, 0);
496
497 error = vn_make_ops("portfs", port_vnodeops_template, &port_vnodeops);
498 if (error) {
499 vfs_freevfsops(port_vfsops);
500 cmn_err(CE_WARN, "port init: bad vnode ops");
501 return (error);
502 }
503
504 mutex_init(&port_control.pc_mutex, NULL, MUTEX_DEFAULT, NULL);
505 port_control.pc_nents = 0; /* number of active ports */
506
507 /* create kmem_cache for port event structures */
508 port_control.pc_cache = kmem_cache_create("port_cache",
509 sizeof (port_kevent_t), 0, NULL, NULL, NULL, NULL, NULL, 0);
510
511 port_kstat_init(); /* init port kstats */
512 return (mod_install(&modlinkage));
513 }
514
515 int
516 _info(struct modinfo *modinfop)
517 {
518 return (mod_info(&modlinkage, modinfop));
519 }
520
521 /*
522 * System call wrapper for all port related system calls from 32-bit programs.
523 */
524 #ifdef _SYSCALL32_IMPL
525 static int64_t
526 portfs32(uint32_t opcode, int32_t a0, uint32_t a1, uint32_t a2, uint32_t a3,
527 uint32_t a4)
528 {
529 int64_t error;
530
531 switch (opcode & PORT_CODE_MASK) {
532 case PORT_GET:
533 error = portfs(PORT_GET, a0, a1, (int)a2, (int)a3, a4);
534 break;
535 case PORT_SENDN:
536 error = portfs(opcode, (uint32_t)a0, a1, a2, a3, a4);
537 break;
538 default:
539 error = portfs(opcode, a0, a1, a2, a3, a4);
540 break;
541 }
542 return (error);
543 }
544 #endif /* _SYSCALL32_IMPL */
545
546 /*
547 * System entry point for port functions.
548 * a0 is a port file descriptor (except for PORT_SENDN and PORT_CREATE).
549 * The libc uses PORT_SYS_NOPORT in functions which do not deliver a
550 * port file descriptor as first argument.
551 */
552 static int64_t
553 portfs(int opcode, uintptr_t a0, uintptr_t a1, uintptr_t a2, uintptr_t a3,
554 uintptr_t a4)
555 {
556 rval_t r;
557 port_t *pp;
558 int error = 0;
559 uint_t nget;
560 file_t *fp;
561 port_gettimer_t port_timer;
562
563 r.r_vals = 0;
564 if (opcode & PORT_SYS_NOPORT) {
565 opcode &= PORT_CODE_MASK;
566 if (opcode == PORT_SENDN) {
567 error = port_sendn((int *)a0, (int *)a1, (uint_t)a2,
568 (int)a3, (void *)a4, (uint_t *)&r.r_val1);
569 if (error && (error != EIO))
570 return ((int64_t)set_errno(error));
571 return (r.r_vals);
572 }
573
574 if (opcode == PORT_CREATE) {
575 error = port_create(&r.r_val1);
576 if (error)
577 return ((int64_t)set_errno(error));
578 return (r.r_vals);
579 }
580 }
581
582 /* opcodes using port as first argument (a0) */
583
584 if ((fp = getf((int)a0)) == NULL)
585 return ((uintptr_t)set_errno(EBADF));
586
587 if (fp->f_vnode->v_type != VPORT) {
588 releasef((int)a0);
589 return ((uintptr_t)set_errno(EBADFD));
590 }
591
592 pp = VTOEP(fp->f_vnode);
593
594 switch (opcode & PORT_CODE_MASK) {
595 case PORT_GET:
596 {
597 /* see PORT_GETN description */
598 struct timespec timeout;
599
600 port_timer.pgt_flags = PORTGET_ONE;
601 port_timer.pgt_loop = 0;
602 port_timer.pgt_rqtp = NULL;
603 if (a4 != NULL) {
604 port_timer.pgt_timeout = &timeout;
605 timeout.tv_sec = (time_t)a2;
606 timeout.tv_nsec = (long)a3;
607 } else {
608 port_timer.pgt_timeout = NULL;
609 }
610 do {
611 nget = 1;
612 error = port_getn(pp, (port_event_t *)a1, 1,
613 (uint_t *)&nget, &port_timer);
614 } while (nget == 0 && error == 0 && port_timer.pgt_loop);
615 break;
616 }
617 case PORT_GETN:
618 {
619 /*
620 * port_getn() can only retrieve own or shareable events from
621 * other processes. The port_getn() function remains in the
622 * kernel until own or shareable events are available or the
623 * timeout elapses.
624 */
625 port_timer.pgt_flags = 0;
626 port_timer.pgt_loop = 0;
627 port_timer.pgt_rqtp = NULL;
628 port_timer.pgt_timeout = (struct timespec *)a4;
629 do {
630 nget = a3;
631 error = port_getn(pp, (port_event_t *)a1, (uint_t)a2,
632 (uint_t *)&nget, &port_timer);
633 } while (nget == 0 && error == 0 && port_timer.pgt_loop);
634 r.r_val1 = nget;
635 r.r_val2 = error;
636 releasef((int)a0);
637 if (error && error != ETIME)
638 return ((int64_t)set_errno(error));
639 return (r.r_vals);
640 }
641 case PORT_ASSOCIATE:
642 {
643 switch ((int)a1) {
644 case PORT_SOURCE_FD:
645 error = port_associate_fd(pp, (int)a1, (uintptr_t)a2,
646 (int)a3, (void *)a4);
647 break;
648 case PORT_SOURCE_FILE:
649 error = port_associate_fop(pp, (int)a1, (uintptr_t)a2,
650 (int)a3, (void *)a4);
651 break;
652 default:
653 error = EINVAL;
654 break;
655 }
656 break;
657 }
658 case PORT_SEND:
659 {
660 /* user-defined events */
661 error = port_send(pp, PORT_SOURCE_USER, (int)a1, (void *)a2);
662 break;
663 }
664 case PORT_DISPATCH:
665 {
666 /*
667 * library events, blocking
668 * Only events of type PORT_SOURCE_AIO or PORT_SOURCE_MQ
669 * are currently allowed.
670 */
671 if ((int)a1 != PORT_SOURCE_AIO && (int)a1 != PORT_SOURCE_MQ) {
672 error = EINVAL;
673 break;
674 }
675 error = port_dispatch_event(pp, (int)opcode, (int)a1, (int)a2,
676 (uintptr_t)a3, (void *)a4);
677 break;
678 }
679 case PORT_DISSOCIATE:
680 {
681 switch ((int)a1) {
682 case PORT_SOURCE_FD:
683 error = port_dissociate_fd(pp, (uintptr_t)a2);
684 break;
685 case PORT_SOURCE_FILE:
686 error = port_dissociate_fop(pp, (uintptr_t)a2);
687 break;
688 default:
689 error = EINVAL;
690 break;
691 }
692 break;
693 }
694 case PORT_ALERT:
695 {
696 if ((int)a2) /* a2 = events */
697 error = port_alert(pp, (int)a1, (int)a2, (void *)a3);
698 else
699 port_remove_alert(&pp->port_queue);
700 break;
701 }
702 default:
703 error = EINVAL;
704 break;
705 }
706
707 releasef((int)a0);
708 if (error)
709 return ((int64_t)set_errno(error));
710 return (r.r_vals);
711 }
712
713 /*
714 * System call to create a port.
715 *
716 * The port_create() function creates a vnode of type VPORT per port.
717 * The port control data is associated with the vnode as vnode private data.
718 * The port_create() function returns an event port file descriptor.
719 */
720 static int
721 port_create(int *fdp)
722 {
723 port_t *pp;
724 vnode_t *vp;
725 struct file *fp;
726 proc_t *p = curproc;
727
728 /* initialize vnode and port private data */
729 pp = kmem_zalloc(sizeof (port_t), KM_SLEEP);
730
731 pp->port_vnode = vn_alloc(KM_SLEEP);
732 vp = EPTOV(pp);
733 vn_setops(vp, port_vnodeops);
734 vp->v_type = VPORT;
735 vp->v_vfsp = &port_vfs;
736 vp->v_data = (caddr_t)pp;
737
738 mutex_enter(&port_control.pc_mutex);
739 /*
740 * Retrieve the maximal number of event ports allowed per system from
741 * the resource control: project.port-max-ids.
742 */
743 mutex_enter(&p->p_lock);
744 if (rctl_test(rc_project_portids, p->p_task->tk_proj->kpj_rctls, p,
745 port_control.pc_nents + 1, RCA_SAFE) & RCT_DENY) {
746 mutex_exit(&p->p_lock);
747 vn_free(vp);
748 kmem_free(pp, sizeof (port_t));
749 mutex_exit(&port_control.pc_mutex);
750 return (EAGAIN);
751 }
752
753 /*
754 * Retrieve the maximal number of events allowed per port from
755 * the resource control: process.port-max-events.
756 */
757 pp->port_max_events = rctl_enforced_value(rc_process_portev,
758 p->p_rctls, p);
759 mutex_exit(&p->p_lock);
760
761 /* allocate a new user file descriptor and a file structure */
762 if (falloc(vp, 0, &fp, fdp)) {
763 /*
764 * If the file table is full, free allocated resources.
765 */
766 vn_free(vp);
767 kmem_free(pp, sizeof (port_t));
768 mutex_exit(&port_control.pc_mutex);
769 return (EMFILE);
770 }
771
772 mutex_exit(&fp->f_tlock);
773
774 pp->port_fd = *fdp;
775 port_control.pc_nents++;
776 p->p_portcnt++;
777 port_kstat.pks_ports.value.ui32++;
778 mutex_exit(&port_control.pc_mutex);
779
780 /* initializes port private data */
781 port_init(pp);
782 /* set user file pointer */
783 setf(*fdp, fp);
784 return (0);
785 }
786
787 /*
788 * port_init() initializes event port specific data
789 */
790 static void
791 port_init(port_t *pp)
792 {
793 port_queue_t *portq;
794 port_ksource_t *pks;
795
796 mutex_init(&pp->port_mutex, NULL, MUTEX_DEFAULT, NULL);
797 portq = &pp->port_queue;
798 mutex_init(&portq->portq_mutex, NULL, MUTEX_DEFAULT, NULL);
799 pp->port_flags |= PORT_INIT;
800
801 /*
802 * If it is not enough memory available to satisfy a user
803 * request using a single port_getn() call then port_getn()
804 * will reduce the size of the list to PORT_MAX_LIST.
805 */
806 pp->port_max_list = port_max_list;
807
808 /* Set timestamp entries required for fstat(2) requests */
809 gethrestime(&pp->port_ctime);
810 pp->port_uid = crgetuid(curproc->p_cred);
811 pp->port_gid = crgetgid(curproc->p_cred);
812
813 /* initialize port queue structs */
814 list_create(&portq->portq_list, sizeof (port_kevent_t),
815 offsetof(port_kevent_t, portkev_node));
816 list_create(&portq->portq_get_list, sizeof (port_kevent_t),
817 offsetof(port_kevent_t, portkev_node));
818 portq->portq_flags = 0;
819 pp->port_pid = curproc->p_pid;
820
821 /* Allocate cache skeleton for PORT_SOURCE_FD events */
822 portq->portq_pcp = kmem_zalloc(sizeof (port_fdcache_t), KM_SLEEP);
823 mutex_init(&portq->portq_pcp->pc_lock, NULL, MUTEX_DEFAULT, NULL);
824
825 /*
826 * Allocate cache skeleton for association of event sources.
827 */
828 mutex_init(&portq->portq_source_mutex, NULL, MUTEX_DEFAULT, NULL);
829 portq->portq_scache = kmem_zalloc(
830 PORT_SCACHE_SIZE * sizeof (port_source_t *), KM_SLEEP);
831
832 /*
833 * pre-associate some kernel sources with this port.
834 * The pre-association is required to create port_source_t
835 * structures for object association.
836 * Some sources can not get associated with a port before the first
837 * object association is requested. Another reason to pre_associate
838 * a particular source with a port is because of performance.
839 */
840
841 for (pks = port_ksource_tab; pks->pks_source != 0; pks++)
842 port_add_ksource_local(pp, pks);
843 }
844
845 /*
846 * The port_add_ksource_local() function is being used to associate
847 * event sources with every new port.
848 * The event sources need to be added to port_ksource_tab[].
849 */
850 static void
851 port_add_ksource_local(port_t *pp, port_ksource_t *pks)
852 {
853 port_source_t *pse;
854 port_source_t **ps;
855
856 mutex_enter(&pp->port_queue.portq_source_mutex);
857 ps = &pp->port_queue.portq_scache[PORT_SHASH(pks->pks_source)];
858 for (pse = *ps; pse != NULL; pse = pse->portsrc_next) {
859 if (pse->portsrc_source == pks->pks_source)
860 break;
861 }
862
863 if (pse == NULL) {
864 /* associate new source with the port */
865 pse = kmem_zalloc(sizeof (port_source_t), KM_SLEEP);
866 pse->portsrc_source = pks->pks_source;
867 pse->portsrc_close = pks->pks_close;
868 pse->portsrc_closearg = pks->pks_closearg;
869 pse->portsrc_cnt = 1;
870
871 pks->pks_portsrc = pse;
872 if (*ps != NULL)
873 pse->portsrc_next = (*ps)->portsrc_next;
874 *ps = pse;
875 }
876 mutex_exit(&pp->port_queue.portq_source_mutex);
877 }
878
879 /*
880 * The port_send() function sends an event of type "source" to a
881 * port. This function is non-blocking. An event can be sent to
882 * a port as long as the number of events per port does not achieve the
883 * maximal allowed number of events. The max. number of events per port is
884 * defined by the resource control process.max-port-events.
885 * This function is used by the port library function port_send()
886 * and port_dispatch(). The port_send(3c) function is part of the
887 * event ports API and submits events of type PORT_SOURCE_USER. The
888 * port_dispatch() function is project private and it is used by library
889 * functions to submit events of other types than PORT_SOURCE_USER
890 * (e.g. PORT_SOURCE_AIO).
891 */
892 static int
893 port_send(port_t *pp, int source, int events, void *user)
894 {
895 port_kevent_t *pev;
896 int error;
897
898 error = port_alloc_event_local(pp, source, PORT_ALLOC_DEFAULT, &pev);
899 if (error)
900 return (error);
901
902 pev->portkev_object = 0;
903 pev->portkev_events = events;
904 pev->portkev_user = user;
905 pev->portkev_callback = NULL;
906 pev->portkev_arg = NULL;
907 pev->portkev_flags = 0;
908
909 port_send_event(pev);
910 return (0);
911 }
912
913 /*
914 * The port_noshare() function returns 0 if the current event was generated
915 * by the same process. Otherwise is returns a value other than 0 and the
916 * event should not be delivered to the current processe.
917 * The port_noshare() function is normally used by the port_dispatch()
918 * function. The port_dispatch() function is project private and can only be
919 * used within the event port project.
920 * Currently the libaio uses the port_dispatch() function to deliver events
921 * of types PORT_SOURCE_AIO.
922 */
923 /* ARGSUSED */
924 static int
925 port_noshare(void *arg, int *events, pid_t pid, int flag, void *evp)
926 {
927 if (flag == PORT_CALLBACK_DEFAULT && curproc->p_pid != pid)
928 return (1);
929 return (0);
930 }
931
932 /*
933 * The port_dispatch_event() function is project private and it is used by
934 * libraries involved in the project to deliver events to the port.
935 * port_dispatch will sleep and wait for enough resources to satisfy the
936 * request, if necessary.
937 * The library can specify if the delivered event is shareable with other
938 * processes (see PORT_SYS_NOSHARE flag).
939 */
940 static int
941 port_dispatch_event(port_t *pp, int opcode, int source, int events,
942 uintptr_t object, void *user)
943 {
944 port_kevent_t *pev;
945 int error;
946
947 error = port_alloc_event_block(pp, source, PORT_ALLOC_DEFAULT, &pev);
948 if (error)
949 return (error);
950
951 pev->portkev_object = object;
952 pev->portkev_events = events;
953 pev->portkev_user = user;
954 pev->portkev_arg = NULL;
955 if (opcode & PORT_SYS_NOSHARE) {
956 pev->portkev_flags = PORT_KEV_NOSHARE;
957 pev->portkev_callback = port_noshare;
958 } else {
959 pev->portkev_flags = 0;
960 pev->portkev_callback = NULL;
961 }
962
963 port_send_event(pev);
964 return (0);
965 }
966
967
968 /*
969 * The port_sendn() function is the kernel implementation of the event
970 * port API function port_sendn(3c).
971 * This function is able to send an event to a list of event ports.
972 */
973 static int
974 port_sendn(int ports[], int errors[], uint_t nent, int events, void *user,
975 uint_t *nget)
976 {
977 port_kevent_t *pev;
978 int errorcnt = 0;
979 int error = 0;
980 int count;
981 int port;
982 int *plist;
983 int *elist = NULL;
984 file_t *fp;
985 port_t *pp;
986
987 if (nent == 0 || nent > port_max_list)
988 return (EINVAL);
989
990 plist = kmem_alloc(nent * sizeof (int), KM_SLEEP);
991 if (copyin((void *)ports, plist, nent * sizeof (int))) {
992 kmem_free(plist, nent * sizeof (int));
993 return (EFAULT);
994 }
995
996 /*
997 * Scan the list for event port file descriptors and send the
998 * attached user event data embedded in a event of type
999 * PORT_SOURCE_USER to every event port in the list.
1000 * If a list entry is not a valid event port then the corresponding
1001 * error code will be stored in the errors[] list with the same
1002 * list offset as in the ports[] list.
1003 */
1004
1005 for (count = 0; count < nent; count++) {
1006 port = plist[count];
1007 if ((fp = getf(port)) == NULL) {
1008 elist = port_errorn(elist, nent, EBADF, count);
1009 errorcnt++;
1010 continue;
1011 }
1012
1013 pp = VTOEP(fp->f_vnode);
1014 if (fp->f_vnode->v_type != VPORT) {
1015 releasef(port);
1016 elist = port_errorn(elist, nent, EBADFD, count);
1017 errorcnt++;
1018 continue;
1019 }
1020
1021 error = port_alloc_event_local(pp, PORT_SOURCE_USER,
1022 PORT_ALLOC_DEFAULT, &pev);
1023 if (error) {
1024 releasef(port);
1025 elist = port_errorn(elist, nent, error, count);
1026 errorcnt++;
1027 continue;
1028 }
1029
1030 pev->portkev_object = 0;
1031 pev->portkev_events = events;
1032 pev->portkev_user = user;
1033 pev->portkev_callback = NULL;
1034 pev->portkev_arg = NULL;
1035 pev->portkev_flags = 0;
1036
1037 port_send_event(pev);
1038 releasef(port);
1039 }
1040 if (errorcnt) {
1041 error = EIO;
1042 if (copyout(elist, (void *)errors, nent * sizeof (int)))
1043 error = EFAULT;
1044 kmem_free(elist, nent * sizeof (int));
1045 }
1046 *nget = nent - errorcnt;
1047 kmem_free(plist, nent * sizeof (int));
1048 return (error);
1049 }
1050
1051 static int *
1052 port_errorn(int *elist, int nent, int error, int index)
1053 {
1054 if (elist == NULL)
1055 elist = kmem_zalloc(nent * sizeof (int), KM_SLEEP);
1056 elist[index] = error;
1057 return (elist);
1058 }
1059
1060 /*
1061 * port_alert()
1062 * The port_alert() funcion is a high priority event and it is always set
1063 * on top of the queue. It is also delivered as single event.
1064 * flags:
1065 * - SET :overwrite current alert data
1066 * - UPDATE:set alert data or return EBUSY if alert mode is already set
1067 *
1068 * - set the ALERT flag
1069 * - wakeup all sleeping threads
1070 */
1071 static int
1072 port_alert(port_t *pp, int flags, int events, void *user)
1073 {
1074 port_queue_t *portq;
1075 portget_t *pgetp;
1076 port_alert_t *pa;
1077
1078 if ((flags & PORT_ALERT_INVALID) == PORT_ALERT_INVALID)
1079 return (EINVAL);
1080
1081 portq = &pp->port_queue;
1082 pa = &portq->portq_alert;
1083 mutex_enter(&portq->portq_mutex);
1084
1085 /* check alert conditions */
1086 if (flags == PORT_ALERT_UPDATE) {
1087 if (portq->portq_flags & PORTQ_ALERT) {
1088 mutex_exit(&portq->portq_mutex);
1089 return (EBUSY);
1090 }
1091 }
1092
1093 /*
1094 * Store alert data in the port to be delivered to threads
1095 * which are using port_get(n) to retrieve events.
1096 */
1097
1098 portq->portq_flags |= PORTQ_ALERT;
1099 pa->portal_events = events; /* alert info */
1100 pa->portal_pid = curproc->p_pid; /* process owner */
1101 pa->portal_object = 0; /* no object */
1102 pa->portal_user = user; /* user alert data */
1103
1104 /* alert and deliver alert data to waiting threads */
1105 pgetp = portq->portq_thread;
1106 if (pgetp == NULL) {
1107 /* no threads waiting for events */
1108 mutex_exit(&portq->portq_mutex);
1109 return (0);
1110 }
1111
1112 /*
1113 * Set waiting threads in alert mode (PORTGET_ALERT)..
1114 * Every thread waiting for events already allocated a portget_t
1115 * structure to sleep on.
1116 * The port alert arguments are stored in the portget_t structure.
1117 * The PORTGET_ALERT flag is set to indicate the thread to return
1118 * immediately with the alert event.
1119 */
1120 do {
1121 if ((pgetp->portget_state & PORTGET_ALERT) == 0) {
1122 pa = &pgetp->portget_alert;
1123 pa->portal_events = events;
1124 pa->portal_object = 0;
1125 pa->portal_user = user;
1126 pgetp->portget_state |= PORTGET_ALERT;
1127 cv_signal(&pgetp->portget_cv);
1128 }
1129 } while ((pgetp = pgetp->portget_next) != portq->portq_thread);
1130 mutex_exit(&portq->portq_mutex);
1131 return (0);
1132 }
1133
1134 /*
1135 * Clear alert state of the port
1136 */
1137 static void
1138 port_remove_alert(port_queue_t *portq)
1139 {
1140 mutex_enter(&portq->portq_mutex);
1141 portq->portq_flags &= ~PORTQ_ALERT;
1142 mutex_exit(&portq->portq_mutex);
1143 }
1144
1145 /*
1146 * The port_getn() function is used to retrieve events from a port.
1147 *
1148 * The port_getn() function returns immediately if there are enough events
1149 * available in the port to satisfy the request or if the port is in alert
1150 * mode (see port_alert(3c)).
1151 * The timeout argument of port_getn(3c) -which is embedded in the
1152 * port_gettimer_t structure- specifies if the system call should block or if it
1153 * should return immediately depending on the number of events available.
1154 * This function is internally used by port_getn(3c) as well as by
1155 * port_get(3c).
1156 */
1157 static int
1158 port_getn(port_t *pp, port_event_t *uevp, uint_t max, uint_t *nget,
1159 port_gettimer_t *pgt)
1160 {
1161 port_queue_t *portq;
1162 port_kevent_t *pev;
1163 port_kevent_t *lev;
1164 int error = 0;
1165 uint_t nmax;
1166 uint_t nevents;
1167 uint_t eventsz;
1168 port_event_t *kevp;
1169 list_t *glist;
1170 uint_t tnent;
1171 int rval;
1172 int blocking = -1;
1173 int timecheck;
1174 int flag;
1175 timespec_t rqtime;
1176 timespec_t *rqtp = NULL;
1177 portget_t *pgetp;
1178 void *results;
1179 model_t model = get_udatamodel();
1180
1181 flag = pgt->pgt_flags;
1182
1183 if (*nget > max && max > 0)
1184 return (EINVAL);
1185
1186 portq = &pp->port_queue;
1187 mutex_enter(&portq->portq_mutex);
1188 if (max == 0) {
1189 /*
1190 * Return number of objects with events.
1191 * The port_block() call is required to synchronize this
1192 * thread with another possible thread, which could be
1193 * retrieving events from the port queue.
1194 */
1195 port_block(portq);
1196 /*
1197 * Check if a second thread is currently retrieving events
1198 * and it is using the temporary event queue.
1199 */
1200 if (portq->portq_tnent) {
1201 /* put remaining events back to the port queue */
1202 port_push_eventq(portq);
1203 }
1204 *nget = portq->portq_nent;
1205 port_unblock(portq);
1206 mutex_exit(&portq->portq_mutex);
1207 return (0);
1208 }
1209
1210 if (uevp == NULL) {
1211 mutex_exit(&portq->portq_mutex);
1212 return (EFAULT);
1213 }
1214 if (*nget == 0) { /* no events required */
1215 mutex_exit(&portq->portq_mutex);
1216 return (0);
1217 }
1218
1219 /* port is being closed ... */
1220 if (portq->portq_flags & PORTQ_CLOSE) {
1221 mutex_exit(&portq->portq_mutex);
1222 return (EBADFD);
1223 }
1224
1225 /* return immediately if port in alert mode */
1226 if (portq->portq_flags & PORTQ_ALERT) {
1227 error = port_get_alert(&portq->portq_alert, uevp);
1228 if (error == 0)
1229 *nget = 1;
1230 mutex_exit(&portq->portq_mutex);
1231 return (error);
1232 }
1233
1234 portq->portq_thrcnt++;
1235
1236 /*
1237 * Now check if the completed events satisfy the
1238 * "wait" requirements of the current thread:
1239 */
1240
1241 if (pgt->pgt_loop) {
1242 /*
1243 * loop entry of same thread
1244 * pgt_loop is set when the current thread returns
1245 * prematurely from this function. That could happen
1246 * when a port is being shared between processes and
1247 * this thread could not find events to return.
1248 * It is not allowed to a thread to retrieve non-shareable
1249 * events generated in other processes.
1250 * PORTQ_WAIT_EVENTS is set when a thread already
1251 * checked the current event queue and no new events
1252 * are added to the queue.
1253 */
1254 if (((portq->portq_flags & PORTQ_WAIT_EVENTS) == 0) &&
1255 (portq->portq_nent >= *nget)) {
1256 /* some new events arrived ...check them */
1257 goto portnowait;
1258 }
1259 rqtp = pgt->pgt_rqtp;
1260 timecheck = pgt->pgt_timecheck;
1261 pgt->pgt_flags |= PORTGET_WAIT_EVENTS;
1262 } else {
1263 /* check if enough events are available ... */
1264 if (portq->portq_nent >= *nget)
1265 goto portnowait;
1266 /*
1267 * There are not enough events available to satisfy
1268 * the request, check timeout value and wait for
1269 * incoming events.
1270 */
1271 error = port_get_timeout(pgt->pgt_timeout, &rqtime, &rqtp,
1272 &blocking, flag);
1273 if (error) {
1274 port_check_return_cond(portq);
1275 mutex_exit(&portq->portq_mutex);
1276 return (error);
1277 }
1278
1279 if (blocking == 0) /* don't block, check fired events */
1280 goto portnowait;
1281
1282 if (rqtp != NULL) {
1283 timespec_t now;
1284 timecheck = timechanged;
1285 gethrestime(&now);
1286 timespecadd(rqtp, &now);
1287 }
1288 }
1289
1290 /* enqueue thread in the list of waiting threads */
1291 pgetp = port_queue_thread(portq, *nget);
1292
1293
1294 /* Wait here until return conditions met */
1295 for (;;) {
1296 if (pgetp->portget_state & PORTGET_ALERT) {
1297 /* reap alert event and return */
1298 error = port_get_alert(&pgetp->portget_alert, uevp);
1299 if (error)
1300 *nget = 0;
1301 else
1302 *nget = 1;
1303 port_dequeue_thread(&pp->port_queue, pgetp);
1304 portq->portq_thrcnt--;
1305 mutex_exit(&portq->portq_mutex);
1306 return (error);
1307 }
1308
1309 /*
1310 * Check if some other thread is already retrieving
1311 * events (portq_getn > 0).
1312 */
1313
1314 if ((portq->portq_getn == 0) &&
1315 ((portq)->portq_nent >= *nget) &&
1316 (!((pgt)->pgt_flags & PORTGET_WAIT_EVENTS) ||
1317 !((portq)->portq_flags & PORTQ_WAIT_EVENTS)))
1318 break;
1319
1320 if (portq->portq_flags & PORTQ_CLOSE) {
1321 error = EBADFD;
1322 break;
1323 }
1324
1325 rval = cv_waituntil_sig(&pgetp->portget_cv, &portq->portq_mutex,
1326 rqtp, timecheck);
1327
1328 if (rval <= 0) {
1329 error = (rval == 0) ? EINTR : ETIME;
1330 break;
1331 }
1332 }
1333
1334 /* take thread out of the wait queue */
1335 port_dequeue_thread(portq, pgetp);
1336
1337 if (error != 0 && (error == EINTR || error == EBADFD ||
1338 (error == ETIME && flag))) {
1339 /* return without events */
1340 port_check_return_cond(portq);
1341 mutex_exit(&portq->portq_mutex);
1342 return (error);
1343 }
1344
1345 portnowait:
1346 /*
1347 * Move port event queue to a temporary event queue .
1348 * New incoming events will be continue be posted to the event queue
1349 * and they will not be considered by the current thread.
1350 * The idea is to avoid lock contentions or an often locking/unlocking
1351 * of the port queue mutex. The contention and performance degradation
1352 * could happen because:
1353 * a) incoming events use the port queue mutex to enqueue new events and
1354 * b) before the event can be delivered to the application it is
1355 * necessary to notify the event sources about the event delivery.
1356 * Sometimes the event sources can require a long time to return and
1357 * the queue mutex would block incoming events.
1358 * During this time incoming events (port_send_event()) do not need
1359 * to awake threads waiting for events. Before the current thread
1360 * returns it will check the conditions to awake other waiting threads.
1361 */
1362 portq->portq_getn++; /* number of threads retrieving events */
1363 port_block(portq); /* block other threads here */
1364 nmax = max < portq->portq_nent ? max : portq->portq_nent;
1365
1366 if (portq->portq_tnent) {
1367 /*
1368 * Move remaining events from previous thread back to the
1369 * port event queue.
1370 */
1371 port_push_eventq(portq);
1372 }
1373 /* move port event queue to a temporary queue */
1374 list_move_tail(&portq->portq_get_list, &portq->portq_list);
1375 glist = &portq->portq_get_list; /* use temporary event queue */
1376 tnent = portq->portq_nent; /* get current number of events */
1377 portq->portq_nent = 0; /* no events in the port event queue */
1378 portq->portq_flags |= PORTQ_WAIT_EVENTS; /* detect incoming events */
1379 mutex_exit(&portq->portq_mutex); /* event queue can be reused now */
1380
1381 if (model == DATAMODEL_NATIVE) {
1382 eventsz = sizeof (port_event_t);
1383 kevp = kmem_alloc(eventsz * nmax, KM_NOSLEEP);
1384 if (kevp == NULL) {
1385 if (nmax > pp->port_max_list)
1386 nmax = pp->port_max_list;
1387 kevp = kmem_alloc(eventsz * nmax, KM_SLEEP);
1388 }
1389 results = kevp;
1390 lev = NULL; /* start with first event in the queue */
1391 for (nevents = 0; nevents < nmax; ) {
1392 pev = port_get_kevent(glist, lev);
1393 if (pev == NULL) /* no more events available */
1394 break;
1395 if (pev->portkev_flags & PORT_KEV_FREE) {
1396 /* Just discard event */
1397 list_remove(glist, pev);
1398 pev->portkev_flags &= ~(PORT_CLEANUP_DONE);
1399 if (PORT_FREE_EVENT(pev))
1400 port_free_event_local(pev, 0);
1401 tnent--;
1402 continue;
1403 }
1404
1405 /* move event data to copyout list */
1406 if (port_copy_event(&kevp[nevents], pev, glist)) {
1407 /*
1408 * Event can not be delivered to the
1409 * current process.
1410 */
1411 if (lev != NULL)
1412 list_insert_after(glist, lev, pev);
1413 else
1414 list_insert_head(glist, pev);
1415 lev = pev; /* last checked event */
1416 } else {
1417 nevents++; /* # of events ready */
1418 }
1419 }
1420 #ifdef _SYSCALL32_IMPL
1421 } else {
1422 port_event32_t *kevp32;
1423
1424 eventsz = sizeof (port_event32_t);
1425 kevp32 = kmem_alloc(eventsz * nmax, KM_NOSLEEP);
1426 if (kevp32 == NULL) {
1427 if (nmax > pp->port_max_list)
1428 nmax = pp->port_max_list;
1429 kevp32 = kmem_alloc(eventsz * nmax, KM_SLEEP);
1430 }
1431 results = kevp32;
1432 lev = NULL; /* start with first event in the queue */
1433 for (nevents = 0; nevents < nmax; ) {
1434 pev = port_get_kevent(glist, lev);
1435 if (pev == NULL) /* no more events available */
1436 break;
1437 if (pev->portkev_flags & PORT_KEV_FREE) {
1438 /* Just discard event */
1439 list_remove(glist, pev);
1440 pev->portkev_flags &= ~(PORT_CLEANUP_DONE);
1441 if (PORT_FREE_EVENT(pev))
1442 port_free_event_local(pev, 0);
1443 tnent--;
1444 continue;
1445 }
1446
1447 /* move event data to copyout list */
1448 if (port_copy_event32(&kevp32[nevents], pev, glist)) {
1449 /*
1450 * Event can not be delivered to the
1451 * current process.
1452 */
1453 if (lev != NULL)
1454 list_insert_after(glist, lev, pev);
1455 else
1456 list_insert_head(glist, pev);
1457 lev = pev; /* last checked event */
1458 } else {
1459 nevents++; /* # of events ready */
1460 }
1461 }
1462 #endif /* _SYSCALL32_IMPL */
1463 }
1464
1465 /*
1466 * Remember number of remaining events in the temporary event queue.
1467 */
1468 portq->portq_tnent = tnent - nevents;
1469
1470 /*
1471 * Work to do before return :
1472 * - push list of remaining events back to the top of the standard
1473 * port queue.
1474 * - if this is the last thread calling port_get(n) then wakeup the
1475 * thread waiting on close(2).
1476 * - check for a deferred cv_signal from port_send_event() and wakeup
1477 * the sleeping thread.
1478 */
1479
1480 mutex_enter(&portq->portq_mutex);
1481 port_unblock(portq);
1482 if (portq->portq_tnent) {
1483 /*
1484 * move remaining events in the temporary event queue back
1485 * to the port event queue
1486 */
1487 port_push_eventq(portq);
1488 }
1489 portq->portq_getn--; /* update # of threads retrieving events */
1490 if (--portq->portq_thrcnt == 0) { /* # of threads waiting ... */
1491 /* Last thread => check close(2) conditions ... */
1492 if (portq->portq_flags & PORTQ_CLOSE) {
1493 cv_signal(&portq->portq_closecv);
1494 mutex_exit(&portq->portq_mutex);
1495 kmem_free(results, eventsz * nmax);
1496 /* do not copyout events */
1497 *nget = 0;
1498 return (EBADFD);
1499 }
1500 } else if (portq->portq_getn == 0) {
1501 /*
1502 * no other threads retrieving events ...
1503 * check wakeup conditions of sleeping threads
1504 */
1505 if ((portq->portq_thread != NULL) &&
1506 (portq->portq_nent >= portq->portq_nget))
1507 cv_signal(&portq->portq_thread->portget_cv);
1508 }
1509
1510 /*
1511 * Check PORTQ_POLLIN here because the current thread set temporarily
1512 * the number of events in the queue to zero.
1513 */
1514 if (portq->portq_flags & PORTQ_POLLIN) {
1515 portq->portq_flags &= ~PORTQ_POLLIN;
1516 mutex_exit(&portq->portq_mutex);
1517 pollwakeup(&pp->port_pollhd, POLLIN);
1518 } else {
1519 mutex_exit(&portq->portq_mutex);
1520 }
1521
1522 /* now copyout list of user event structures to user space */
1523 if (nevents) {
1524 if (copyout(results, uevp, nevents * eventsz))
1525 error = EFAULT;
1526 }
1527 kmem_free(results, eventsz * nmax);
1528
1529 if (nevents == 0 && error == 0 && pgt->pgt_loop == 0 && blocking != 0) {
1530 /* no events retrieved: check loop conditions */
1531 if (blocking == -1) {
1532 /* no timeout checked */
1533 error = port_get_timeout(pgt->pgt_timeout,
1534 &pgt->pgt_rqtime, &rqtp, &blocking, flag);
1535 if (error) {
1536 *nget = nevents;
1537 return (error);
1538 }
1539 if (rqtp != NULL) {
1540 timespec_t now;
1541 pgt->pgt_timecheck = timechanged;
1542 gethrestime(&now);
1543 timespecadd(&pgt->pgt_rqtime, &now);
1544 }
1545 pgt->pgt_rqtp = rqtp;
1546 } else {
1547 /* timeout already checked -> remember values */
1548 pgt->pgt_rqtp = rqtp;
1549 if (rqtp != NULL) {
1550 pgt->pgt_timecheck = timecheck;
1551 pgt->pgt_rqtime = *rqtp;
1552 }
1553 }
1554 if (blocking)
1555 /* timeout remaining */
1556 pgt->pgt_loop = 1;
1557 }
1558
1559 /* set number of user event structures completed */
1560 *nget = nevents;
1561 return (error);
1562 }
1563
1564 /*
1565 * 1. copy kernel event structure to user event structure.
1566 * 2. PORT_KEV_WIRED event structures will be reused by the "source"
1567 * 3. Remove PORT_KEV_DONEQ flag (event removed from the event queue)
1568 * 4. Other types of event structures can be delivered back to the port cache
1569 * (port_free_event_local()).
1570 * 5. The event source callback function is the last opportunity for the
1571 * event source to update events, to free local resources associated with
1572 * the event or to deny the delivery of the event.
1573 */
1574 static int
1575 port_copy_event(port_event_t *puevp, port_kevent_t *pkevp, list_t *list)
1576 {
1577 int free_event = 0;
1578 int flags;
1579 int error;
1580
1581 puevp->portev_source = pkevp->portkev_source;
1582 puevp->portev_object = pkevp->portkev_object;
1583 puevp->portev_user = pkevp->portkev_user;
1584 puevp->portev_events = pkevp->portkev_events;
1585
1586 /* remove event from the queue */
1587 list_remove(list, pkevp);
1588
1589 /*
1590 * Events of type PORT_KEV_WIRED remain allocated by the
1591 * event source.
1592 */
1593 flags = pkevp->portkev_flags;
1594 if (pkevp->portkev_flags & PORT_KEV_WIRED)
1595 pkevp->portkev_flags &= ~PORT_KEV_DONEQ;
1596 else
1597 free_event = 1;
1598
1599 if (pkevp->portkev_callback) {
1600 error = (*pkevp->portkev_callback)(pkevp->portkev_arg,
1601 &puevp->portev_events, pkevp->portkev_pid,
1602 PORT_CALLBACK_DEFAULT, pkevp);
1603
1604 if (error) {
1605 /*
1606 * Event can not be delivered.
1607 * Caller must reinsert the event into the queue.
1608 */
1609 pkevp->portkev_flags = flags;
1610 return (error);
1611 }
1612 }
1613 if (free_event)
1614 port_free_event_local(pkevp, 0);
1615 return (0);
1616 }
1617
1618 #ifdef _SYSCALL32_IMPL
1619 /*
1620 * 1. copy kernel event structure to user event structure.
1621 * 2. PORT_KEV_WIRED event structures will be reused by the "source"
1622 * 3. Remove PORT_KEV_DONEQ flag (event removed from the event queue)
1623 * 4. Other types of event structures can be delivered back to the port cache
1624 * (port_free_event_local()).
1625 * 5. The event source callback function is the last opportunity for the
1626 * event source to update events, to free local resources associated with
1627 * the event or to deny the delivery of the event.
1628 */
1629 static int
1630 port_copy_event32(port_event32_t *puevp, port_kevent_t *pkevp, list_t *list)
1631 {
1632 int free_event = 0;
1633 int error;
1634 int flags;
1635
1636 puevp->portev_source = pkevp->portkev_source;
1637 puevp->portev_object = (daddr32_t)pkevp->portkev_object;
1638 puevp->portev_user = (caddr32_t)(uintptr_t)pkevp->portkev_user;
1639 puevp->portev_events = pkevp->portkev_events;
1640
1641 /* remove event from the queue */
1642 list_remove(list, pkevp);
1643
1644 /*
1645 * Events if type PORT_KEV_WIRED remain allocated by the
1646 * sub-system (source).
1647 */
1648
1649 flags = pkevp->portkev_flags;
1650 if (pkevp->portkev_flags & PORT_KEV_WIRED)
1651 pkevp->portkev_flags &= ~PORT_KEV_DONEQ;
1652 else
1653 free_event = 1;
1654
1655 if (pkevp->portkev_callback != NULL) {
1656 error = (*pkevp->portkev_callback)(pkevp->portkev_arg,
1657 &puevp->portev_events, pkevp->portkev_pid,
1658 PORT_CALLBACK_DEFAULT, pkevp);
1659 if (error) {
1660 /*
1661 * Event can not be delivered.
1662 * Caller must reinsert the event into the queue.
1663 */
1664 pkevp->portkev_flags = flags;
1665 return (error);
1666 }
1667 }
1668 if (free_event)
1669 port_free_event_local(pkevp, 0);
1670 return (0);
1671 }
1672 #endif /* _SYSCALL32_IMPL */
1673
1674 /*
1675 * copyout alert event.
1676 */
1677 static int
1678 port_get_alert(port_alert_t *pa, port_event_t *uevp)
1679 {
1680 model_t model = get_udatamodel();
1681
1682 /* copyout alert event structures to user space */
1683 if (model == DATAMODEL_NATIVE) {
1684 port_event_t uev;
1685 uev.portev_source = PORT_SOURCE_ALERT;
1686 uev.portev_object = pa->portal_object;
1687 uev.portev_events = pa->portal_events;
1688 uev.portev_user = pa->portal_user;
1689 if (copyout(&uev, uevp, sizeof (port_event_t)))
1690 return (EFAULT);
1691 #ifdef _SYSCALL32_IMPL
1692 } else {
1693 port_event32_t uev32;
1694 uev32.portev_source = PORT_SOURCE_ALERT;
1695 uev32.portev_object = (daddr32_t)pa->portal_object;
1696 uev32.portev_events = pa->portal_events;
1697 uev32.portev_user = (daddr32_t)(uintptr_t)pa->portal_user;
1698 if (copyout(&uev32, uevp, sizeof (port_event32_t)))
1699 return (EFAULT);
1700 #endif /* _SYSCALL32_IMPL */
1701 }
1702 return (0);
1703 }
1704
1705 /*
1706 * Check return conditions :
1707 * - pending port close(2)
1708 * - threads waiting for events
1709 */
1710 static void
1711 port_check_return_cond(port_queue_t *portq)
1712 {
1713 ASSERT(MUTEX_HELD(&portq->portq_mutex));
1714 portq->portq_thrcnt--;
1715 if (portq->portq_flags & PORTQ_CLOSE) {
1716 if (portq->portq_thrcnt == 0)
1717 cv_signal(&portq->portq_closecv);
1718 else
1719 cv_signal(&portq->portq_thread->portget_cv);
1720 }
1721 }
1722
1723 /*
1724 * The port_get_kevent() function returns
1725 * - the event located at the head of the queue if 'last' pointer is NULL
1726 * - the next event after the event pointed by 'last'
1727 * The caller of this function is responsible for the integrity of the queue
1728 * in use:
1729 * - port_getn() is using a temporary queue protected with port_block().
1730 * - port_close_events() is working on the global event queue and protects
1731 * the queue with portq->portq_mutex.
1732 */
1733 port_kevent_t *
1734 port_get_kevent(list_t *list, port_kevent_t *last)
1735 {
1736 if (last == NULL)
1737 return (list_head(list));
1738 else
1739 return (list_next(list, last));
1740 }
1741
1742 /*
1743 * The port_get_timeout() function gets the timeout data from user space
1744 * and converts that info into a corresponding internal representation.
1745 * The kerneldata flag means that the timeout data is already loaded.
1746 */
1747 static int
1748 port_get_timeout(timespec_t *timeout, timespec_t *rqtime, timespec_t **rqtp,
1749 int *blocking, int kerneldata)
1750 {
1751 model_t model = get_udatamodel();
1752
1753 *rqtp = NULL;
1754 if (timeout == NULL) {
1755 *blocking = 1;
1756 return (0);
1757 }
1758
1759 if (kerneldata) {
1760 *rqtime = *timeout;
1761 } else {
1762 if (model == DATAMODEL_NATIVE) {
1763 if (copyin(timeout, rqtime, sizeof (*rqtime)))
1764 return (EFAULT);
1765 #ifdef _SYSCALL32_IMPL
1766 } else {
1767 timespec32_t wait_time_32;
1768 if (copyin(timeout, &wait_time_32,
1769 sizeof (wait_time_32)))
1770 return (EFAULT);
1771 TIMESPEC32_TO_TIMESPEC(rqtime, &wait_time_32);
1772 #endif /* _SYSCALL32_IMPL */
1773 }
1774 }
1775
1776 if (rqtime->tv_sec == 0 && rqtime->tv_nsec == 0) {
1777 *blocking = 0;
1778 return (0);
1779 }
1780
1781 if (rqtime->tv_sec < 0 ||
1782 rqtime->tv_nsec < 0 || rqtime->tv_nsec >= NANOSEC)
1783 return (EINVAL);
1784
1785 *rqtp = rqtime;
1786 *blocking = 1;
1787 return (0);
1788 }
1789
1790 /*
1791 * port_queue_thread()
1792 * Threads requiring more events than available will be put in a wait queue.
1793 * There is a "thread wait queue" per port.
1794 * Threads requiring less events get a higher priority than others and they
1795 * will be awoken first.
1796 */
1797 static portget_t *
1798 port_queue_thread(port_queue_t *portq, uint_t nget)
1799 {
1800 portget_t *pgetp;
1801 portget_t *ttp;
1802 portget_t *htp;
1803
1804 pgetp = kmem_zalloc(sizeof (portget_t), KM_SLEEP);
1805 pgetp->portget_nget = nget;
1806 pgetp->portget_pid = curproc->p_pid;
1807 if (portq->portq_thread == NULL) {
1808 /* first waiting thread */
1809 portq->portq_thread = pgetp;
1810 portq->portq_nget = nget;
1811 pgetp->portget_prev = pgetp;
1812 pgetp->portget_next = pgetp;
1813 return (pgetp);
1814 }
1815
1816 /*
1817 * thread waiting for less events will be set on top of the queue.
1818 */
1819 ttp = portq->portq_thread;
1820 htp = ttp;
1821 for (;;) {
1822 if (nget <= ttp->portget_nget)
1823 break;
1824 if (htp == ttp->portget_next)
1825 break; /* last event */
1826 ttp = ttp->portget_next;
1827 }
1828
1829 /* add thread to the queue */
1830 pgetp->portget_next = ttp;
1831 pgetp->portget_prev = ttp->portget_prev;
1832 ttp->portget_prev->portget_next = pgetp;
1833 ttp->portget_prev = pgetp;
1834 if (portq->portq_thread == ttp)
1835 portq->portq_thread = pgetp;
1836 portq->portq_nget = portq->portq_thread->portget_nget;
1837 return (pgetp);
1838 }
1839
1840 /*
1841 * Take thread out of the queue.
1842 */
1843 static void
1844 port_dequeue_thread(port_queue_t *portq, portget_t *pgetp)
1845 {
1846 if (pgetp->portget_next == pgetp) {
1847 /* last (single) waiting thread */
1848 portq->portq_thread = NULL;
1849 portq->portq_nget = 0;
1850 } else {
1851 pgetp->portget_prev->portget_next = pgetp->portget_next;
1852 pgetp->portget_next->portget_prev = pgetp->portget_prev;
1853 if (portq->portq_thread == pgetp)
1854 portq->portq_thread = pgetp->portget_next;
1855 portq->portq_nget = portq->portq_thread->portget_nget;
1856 }
1857 kmem_free(pgetp, sizeof (portget_t));
1858 }
1859
1860 /*
1861 * Set up event port kstats.
1862 */
1863 static void
1864 port_kstat_init()
1865 {
1866 kstat_t *ksp;
1867 uint_t ndata;
1868
1869 ndata = sizeof (port_kstat) / sizeof (kstat_named_t);
1870 ksp = kstat_create("portfs", 0, "Event Ports", "misc",
1871 KSTAT_TYPE_NAMED, ndata, KSTAT_FLAG_VIRTUAL);
1872 if (ksp) {
1873 ksp->ks_data = &port_kstat;
1874 kstat_install(ksp);
1875 }
1876 }