xref: /dragonfly/sys/kern/sys_pipe.c (revision ae24b5e0)
1 /*
2  * Copyright (c) 1996 John S. Dyson
3  * All rights reserved.
4  *
5  * Redistribution and use in source and binary forms, with or without
6  * modification, are permitted provided that the following conditions
7  * are met:
8  * 1. Redistributions of source code must retain the above copyright
9  *    notice immediately at the beginning of the file, without modification,
10  *    this list of conditions, and the following disclaimer.
11  * 2. Redistributions in binary form must reproduce the above copyright
12  *    notice, this list of conditions and the following disclaimer in the
13  *    documentation and/or other materials provided with the distribution.
14  * 3. Absolutely no warranty of function or purpose is made by the author
15  *    John S. Dyson.
16  * 4. Modifications may be freely made to this file if the above conditions
17  *    are met.
18  *
19  * $FreeBSD: src/sys/kern/sys_pipe.c,v 1.60.2.13 2002/08/05 15:05:15 des Exp $
20  */
21 
22 /*
23  * This file contains a high-performance replacement for the socket-based
24  * pipes scheme originally used in FreeBSD/4.4Lite.  It does not support
25  * all features of sockets, but does do everything that pipes normally
26  * do.
27  */
28 #include <sys/param.h>
29 #include <sys/systm.h>
30 #include <sys/kernel.h>
31 #include <sys/proc.h>
32 #include <sys/fcntl.h>
33 #include <sys/file.h>
34 #include <sys/filedesc.h>
35 #include <sys/filio.h>
36 #include <sys/ttycom.h>
37 #include <sys/stat.h>
38 #include <sys/signalvar.h>
39 #include <sys/sysproto.h>
40 #include <sys/pipe.h>
41 #include <sys/vnode.h>
42 #include <sys/uio.h>
43 #include <sys/event.h>
44 #include <sys/globaldata.h>
45 #include <sys/module.h>
46 #include <sys/malloc.h>
47 #include <sys/sysctl.h>
48 #include <sys/socket.h>
49 #include <sys/kern_syscall.h>
50 
51 #include <vm/vm.h>
52 #include <vm/vm_param.h>
53 #include <sys/lock.h>
54 #include <vm/vm_object.h>
55 #include <vm/vm_kern.h>
56 #include <vm/vm_extern.h>
57 #include <vm/pmap.h>
58 #include <vm/vm_map.h>
59 #include <vm/vm_page.h>
60 #include <vm/vm_zone.h>
61 
62 #include <sys/file2.h>
63 #include <sys/signal2.h>
64 
65 #include <machine/cpufunc.h>
66 
67 /*
68  * interfaces to the outside world
69  */
70 static int pipe_read (struct file *fp, struct uio *uio,
71 		struct ucred *cred, int flags);
72 static int pipe_write (struct file *fp, struct uio *uio,
73 		struct ucred *cred, int flags);
74 static int pipe_close (struct file *fp);
75 static int pipe_shutdown (struct file *fp, int how);
76 static int pipe_kqfilter (struct file *fp, struct knote *kn);
77 static int pipe_stat (struct file *fp, struct stat *sb, struct ucred *cred);
78 static int pipe_ioctl (struct file *fp, u_long cmd, caddr_t data,
79 		struct ucred *cred, struct sysmsg *msg);
80 
81 static struct fileops pipeops = {
82 	.fo_read = pipe_read,
83 	.fo_write = pipe_write,
84 	.fo_ioctl = pipe_ioctl,
85 	.fo_kqfilter = pipe_kqfilter,
86 	.fo_stat = pipe_stat,
87 	.fo_close = pipe_close,
88 	.fo_shutdown = pipe_shutdown
89 };
90 
91 static void	filt_pipedetach(struct knote *kn);
92 static int	filt_piperead(struct knote *kn, long hint);
93 static int	filt_pipewrite(struct knote *kn, long hint);
94 
95 static struct filterops pipe_rfiltops =
96 	{ FILTEROP_ISFD|FILTEROP_MPSAFE, NULL, filt_pipedetach, filt_piperead };
97 static struct filterops pipe_wfiltops =
98 	{ FILTEROP_ISFD|FILTEROP_MPSAFE, NULL, filt_pipedetach, filt_pipewrite };
99 
100 MALLOC_DEFINE(M_PIPE, "pipe", "pipe structures");
101 
102 /*
103  * Default pipe buffer size(s), this can be kind-of large now because pipe
104  * space is pageable.  The pipe code will try to maintain locality of
105  * reference for performance reasons, so small amounts of outstanding I/O
106  * will not wipe the cache.
107  */
108 #define MINPIPESIZE (PIPE_SIZE/3)
109 #define MAXPIPESIZE (2*PIPE_SIZE/3)
110 
111 /*
112  * Limit the number of "big" pipes
113  */
114 #define LIMITBIGPIPES	64
115 #define PIPEQ_MAX_CACHE 16      /* per-cpu pipe structure cache */
116 
117 static int pipe_maxbig = LIMITBIGPIPES;
118 static int pipe_maxcache = PIPEQ_MAX_CACHE;
119 static int pipe_bigcount;
120 static int pipe_nbig;
121 static int pipe_bcache_alloc;
122 static int pipe_bkmem_alloc;
123 static int pipe_rblocked_count;
124 static int pipe_wblocked_count;
125 
126 SYSCTL_NODE(_kern, OID_AUTO, pipe, CTLFLAG_RW, 0, "Pipe operation");
127 SYSCTL_INT(_kern_pipe, OID_AUTO, nbig,
128         CTLFLAG_RD, &pipe_nbig, 0, "number of big pipes allocated");
129 SYSCTL_INT(_kern_pipe, OID_AUTO, bigcount,
130         CTLFLAG_RW, &pipe_bigcount, 0, "number of times pipe expanded");
131 SYSCTL_INT(_kern_pipe, OID_AUTO, rblocked,
132         CTLFLAG_RW, &pipe_rblocked_count, 0, "number of times pipe expanded");
133 SYSCTL_INT(_kern_pipe, OID_AUTO, wblocked,
134         CTLFLAG_RW, &pipe_wblocked_count, 0, "number of times pipe expanded");
135 SYSCTL_INT(_kern_pipe, OID_AUTO, maxcache,
136         CTLFLAG_RW, &pipe_maxcache, 0, "max pipes cached per-cpu");
137 SYSCTL_INT(_kern_pipe, OID_AUTO, maxbig,
138         CTLFLAG_RW, &pipe_maxbig, 0, "max number of big pipes");
139 static int pipe_delay = 5000;	/* 5uS default */
140 SYSCTL_INT(_kern_pipe, OID_AUTO, delay,
141         CTLFLAG_RW, &pipe_delay, 0, "SMP delay optimization in ns");
142 #if !defined(NO_PIPE_SYSCTL_STATS)
143 SYSCTL_INT(_kern_pipe, OID_AUTO, bcache_alloc,
144         CTLFLAG_RW, &pipe_bcache_alloc, 0, "pipe buffer from pcpu cache");
145 SYSCTL_INT(_kern_pipe, OID_AUTO, bkmem_alloc,
146         CTLFLAG_RW, &pipe_bkmem_alloc, 0, "pipe buffer from kmem");
147 #endif
148 
149 /*
150  * Auto-size pipe cache to reduce kmem allocations and frees.
151  */
152 static
153 void
154 pipeinit(void *dummy)
155 {
156 	size_t mbytes = kmem_lim_size();
157 
158 	if (pipe_maxbig == LIMITBIGPIPES) {
159 		if (mbytes >= 7 * 1024)
160 			pipe_maxbig *= 2;
161 		if (mbytes >= 15 * 1024)
162 			pipe_maxbig *= 2;
163 	}
164 	if (pipe_maxcache == PIPEQ_MAX_CACHE) {
165 		if (mbytes >= 7 * 1024)
166 			pipe_maxcache *= 2;
167 		if (mbytes >= 15 * 1024)
168 			pipe_maxcache *= 2;
169 	}
170 }
171 SYSINIT(kmem, SI_BOOT2_MACHDEP, SI_ORDER_ANY, pipeinit, NULL);
172 
173 static void pipeclose (struct pipe *cpipe);
174 static void pipe_free_kmem (struct pipe *cpipe);
175 static int pipe_create (struct pipe **cpipep);
176 static int pipespace (struct pipe *cpipe, int size);
177 
178 static __inline void
179 pipewakeup(struct pipe *cpipe, int dosigio)
180 {
181 	if (dosigio && (cpipe->pipe_state & PIPE_ASYNC) && cpipe->pipe_sigio) {
182 		lwkt_gettoken(&sigio_token);
183 		pgsigio(cpipe->pipe_sigio, SIGIO, 0);
184 		lwkt_reltoken(&sigio_token);
185 	}
186 	KNOTE(&cpipe->pipe_kq.ki_note, 0);
187 }
188 
189 /*
190  * These routines are called before and after a UIO.  The UIO
191  * may block, causing our held tokens to be lost temporarily.
192  *
193  * We use these routines to serialize reads against other reads
194  * and writes against other writes.
195  *
196  * The read token is held on entry so *ipp does not race.
197  */
198 static __inline int
199 pipe_start_uio(struct pipe *cpipe, int *ipp)
200 {
201 	int error;
202 
203 	while (*ipp) {
204 		*ipp = -1;
205 		error = tsleep(ipp, PCATCH, "pipexx", 0);
206 		if (error)
207 			return (error);
208 	}
209 	*ipp = 1;
210 	return (0);
211 }
212 
213 static __inline void
214 pipe_end_uio(struct pipe *cpipe, int *ipp)
215 {
216 	if (*ipp < 0) {
217 		*ipp = 0;
218 		wakeup(ipp);
219 	} else {
220 		KKASSERT(*ipp > 0);
221 		*ipp = 0;
222 	}
223 }
224 
225 /*
226  * The pipe system call for the DTYPE_PIPE type of pipes
227  *
228  * pipe_args(int dummy)
229  *
230  * MPSAFE
231  */
232 int
233 sys_pipe(struct pipe_args *uap)
234 {
235 	return kern_pipe(uap->sysmsg_fds, 0);
236 }
237 
238 int
239 sys_pipe2(struct pipe2_args *uap)
240 {
241 	return kern_pipe(uap->sysmsg_fds, uap->flags);
242 }
243 
244 int
245 kern_pipe(long *fds, int flags)
246 {
247 	struct thread *td = curthread;
248 	struct filedesc *fdp = td->td_proc->p_fd;
249 	struct file *rf, *wf;
250 	struct pipe *rpipe, *wpipe;
251 	int fd1, fd2, error;
252 
253 	rpipe = wpipe = NULL;
254 	if (pipe_create(&rpipe) || pipe_create(&wpipe)) {
255 		pipeclose(rpipe);
256 		pipeclose(wpipe);
257 		return (ENFILE);
258 	}
259 
260 	error = falloc(td->td_lwp, &rf, &fd1);
261 	if (error) {
262 		pipeclose(rpipe);
263 		pipeclose(wpipe);
264 		return (error);
265 	}
266 	fds[0] = fd1;
267 
268 	/*
269 	 * Warning: once we've gotten past allocation of the fd for the
270 	 * read-side, we can only drop the read side via fdrop() in order
271 	 * to avoid races against processes which manage to dup() the read
272 	 * side while we are blocked trying to allocate the write side.
273 	 */
274 	rf->f_type = DTYPE_PIPE;
275 	rf->f_flag = FREAD | FWRITE;
276 	rf->f_ops = &pipeops;
277 	rf->f_data = rpipe;
278 	if (flags & O_NONBLOCK)
279 		rf->f_flag |= O_NONBLOCK;
280 	if (flags & O_CLOEXEC)
281 		fdp->fd_files[fd1].fileflags |= UF_EXCLOSE;
282 
283 	error = falloc(td->td_lwp, &wf, &fd2);
284 	if (error) {
285 		fsetfd(fdp, NULL, fd1);
286 		fdrop(rf);
287 		/* rpipe has been closed by fdrop(). */
288 		pipeclose(wpipe);
289 		return (error);
290 	}
291 	wf->f_type = DTYPE_PIPE;
292 	wf->f_flag = FREAD | FWRITE;
293 	wf->f_ops = &pipeops;
294 	wf->f_data = wpipe;
295 	if (flags & O_NONBLOCK)
296 		wf->f_flag |= O_NONBLOCK;
297 	if (flags & O_CLOEXEC)
298 		fdp->fd_files[fd2].fileflags |= UF_EXCLOSE;
299 
300 	fds[1] = fd2;
301 
302 	rpipe->pipe_slock = kmalloc(sizeof(struct lock),
303 				    M_PIPE, M_WAITOK|M_ZERO);
304 	wpipe->pipe_slock = rpipe->pipe_slock;
305 	rpipe->pipe_peer = wpipe;
306 	wpipe->pipe_peer = rpipe;
307 	lockinit(rpipe->pipe_slock, "pipecl", 0, 0);
308 
309 	/*
310 	 * Once activated the peer relationship remains valid until
311 	 * both sides are closed.
312 	 */
313 	fsetfd(fdp, rf, fd1);
314 	fsetfd(fdp, wf, fd2);
315 	fdrop(rf);
316 	fdrop(wf);
317 
318 	return (0);
319 }
320 
321 /*
322  * Allocate kva for pipe circular buffer, the space is pageable
323  * This routine will 'realloc' the size of a pipe safely, if it fails
324  * it will retain the old buffer.
325  * If it fails it will return ENOMEM.
326  */
327 static int
328 pipespace(struct pipe *cpipe, int size)
329 {
330 	struct vm_object *object;
331 	caddr_t buffer;
332 	int npages, error;
333 
334 	npages = round_page(size) / PAGE_SIZE;
335 	object = cpipe->pipe_buffer.object;
336 
337 	/*
338 	 * [re]create the object if necessary and reserve space for it
339 	 * in the kernel_map.  The object and memory are pageable.  On
340 	 * success, free the old resources before assigning the new
341 	 * ones.
342 	 */
343 	if (object == NULL || object->size != npages) {
344 		object = vm_object_allocate(OBJT_DEFAULT, npages);
345 		buffer = (caddr_t)vm_map_min(&kernel_map);
346 
347 		error = vm_map_find(&kernel_map, object, NULL,
348 				    0, (vm_offset_t *)&buffer, size,
349 				    PAGE_SIZE, TRUE,
350 				    VM_MAPTYPE_NORMAL, VM_SUBSYS_PIPE,
351 				    VM_PROT_ALL, VM_PROT_ALL, 0);
352 
353 		if (error != KERN_SUCCESS) {
354 			vm_object_deallocate(object);
355 			return (ENOMEM);
356 		}
357 		pipe_free_kmem(cpipe);
358 		cpipe->pipe_buffer.object = object;
359 		cpipe->pipe_buffer.buffer = buffer;
360 		cpipe->pipe_buffer.size = size;
361 		++pipe_bkmem_alloc;
362 	} else {
363 		++pipe_bcache_alloc;
364 	}
365 	cpipe->pipe_buffer.rindex = 0;
366 	cpipe->pipe_buffer.windex = 0;
367 	return (0);
368 }
369 
370 /*
371  * Initialize and allocate VM and memory for pipe, pulling the pipe from
372  * our per-cpu cache if possible.  For now make sure it is sized for the
373  * smaller PIPE_SIZE default.
374  */
375 static int
376 pipe_create(struct pipe **cpipep)
377 {
378 	globaldata_t gd = mycpu;
379 	struct pipe *cpipe;
380 	int error;
381 
382 	if ((cpipe = gd->gd_pipeq) != NULL) {
383 		gd->gd_pipeq = cpipe->pipe_peer;
384 		--gd->gd_pipeqcount;
385 		cpipe->pipe_peer = NULL;
386 		cpipe->pipe_wantwcnt = 0;
387 	} else {
388 		cpipe = kmalloc(sizeof(struct pipe), M_PIPE, M_WAITOK|M_ZERO);
389 	}
390 	*cpipep = cpipe;
391 	if ((error = pipespace(cpipe, PIPE_SIZE)) != 0)
392 		return (error);
393 	vfs_timestamp(&cpipe->pipe_ctime);
394 	cpipe->pipe_atime = cpipe->pipe_ctime;
395 	cpipe->pipe_mtime = cpipe->pipe_ctime;
396 	lwkt_token_init(&cpipe->pipe_rlock, "piper");
397 	lwkt_token_init(&cpipe->pipe_wlock, "pipew");
398 	return (0);
399 }
400 
401 static int
402 pipe_read(struct file *fp, struct uio *uio, struct ucred *cred, int fflags)
403 {
404 	struct pipe *rpipe;
405 	struct pipe *wpipe;
406 	int error;
407 	size_t nread = 0;
408 	int nbio;
409 	u_int size;	/* total bytes available */
410 	u_int nsize;	/* total bytes to read */
411 	u_int rindex;	/* contiguous bytes available */
412 	int notify_writer;
413 	int bigread;
414 	int bigcount;
415 
416 	atomic_set_int(&curthread->td_mpflags, TDF_MP_BATCH_DEMARC);
417 
418 	if (uio->uio_resid == 0)
419 		return(0);
420 
421 	/*
422 	 * Setup locks, calculate nbio
423 	 */
424 	rpipe = (struct pipe *)fp->f_data;
425 	wpipe = rpipe->pipe_peer;
426 	lwkt_gettoken(&rpipe->pipe_rlock);
427 
428 	if (fflags & O_FBLOCKING)
429 		nbio = 0;
430 	else if (fflags & O_FNONBLOCKING)
431 		nbio = 1;
432 	else if (fp->f_flag & O_NONBLOCK)
433 		nbio = 1;
434 	else
435 		nbio = 0;
436 
437 	/*
438 	 * Reads are serialized.  Note however that pipe_buffer.buffer and
439 	 * pipe_buffer.size can change out from under us when the number
440 	 * of bytes in the buffer are zero due to the write-side doing a
441 	 * pipespace().
442 	 */
443 	error = pipe_start_uio(rpipe, &rpipe->pipe_rip);
444 	if (error) {
445 		lwkt_reltoken(&rpipe->pipe_rlock);
446 		return (error);
447 	}
448 	notify_writer = 0;
449 
450 	bigread = (uio->uio_resid > 10 * 1024 * 1024);
451 	bigcount = 10;
452 
453 	while (uio->uio_resid) {
454 		/*
455 		 * Don't hog the cpu.
456 		 */
457 		if (bigread && --bigcount == 0) {
458 			lwkt_user_yield();
459 			bigcount = 10;
460 			if (CURSIG(curthread->td_lwp)) {
461 				error = EINTR;
462 				break;
463 			}
464 		}
465 
466 		size = rpipe->pipe_buffer.windex - rpipe->pipe_buffer.rindex;
467 		cpu_lfence();
468 		if (size) {
469 			rindex = rpipe->pipe_buffer.rindex &
470 				 (rpipe->pipe_buffer.size - 1);
471 			nsize = size;
472 			if (nsize > rpipe->pipe_buffer.size - rindex)
473 				nsize = rpipe->pipe_buffer.size - rindex;
474 			nsize = szmin(nsize, uio->uio_resid);
475 
476 			error = uiomove(&rpipe->pipe_buffer.buffer[rindex],
477 					nsize, uio);
478 			if (error)
479 				break;
480 			cpu_mfence();
481 			rpipe->pipe_buffer.rindex += nsize;
482 			nread += nsize;
483 
484 			/*
485 			 * If the FIFO is still over half full just continue
486 			 * and do not try to notify the writer yet.
487 			 */
488 			if (size - nsize >= (rpipe->pipe_buffer.size >> 1)) {
489 				notify_writer = 0;
490 				continue;
491 			}
492 
493 			/*
494 			 * When the FIFO is less then half full notify any
495 			 * waiting writer.  WANTW can be checked while
496 			 * holding just the rlock.
497 			 */
498 			notify_writer = 1;
499 			if ((rpipe->pipe_state & PIPE_WANTW) == 0)
500 				continue;
501 		}
502 
503 		/*
504 		 * If the "write-side" was blocked we wake it up.  This code
505 		 * is reached either when the buffer is completely emptied
506 		 * or if it becomes more then half-empty.
507 		 *
508 		 * Pipe_state can only be modified if both the rlock and
509 		 * wlock are held.
510 		 */
511 		if (rpipe->pipe_state & PIPE_WANTW) {
512 			lwkt_gettoken(&rpipe->pipe_wlock);
513 			if (rpipe->pipe_state & PIPE_WANTW) {
514 				rpipe->pipe_state &= ~PIPE_WANTW;
515 				lwkt_reltoken(&rpipe->pipe_wlock);
516 				wakeup(rpipe);
517 			} else {
518 				lwkt_reltoken(&rpipe->pipe_wlock);
519 			}
520 		}
521 
522 		/*
523 		 * Pick up our copy loop again if the writer sent data to
524 		 * us while we were messing around.
525 		 *
526 		 * On a SMP box poll up to pipe_delay nanoseconds for new
527 		 * data.  Typically a value of 2000 to 4000 is sufficient
528 		 * to eradicate most IPIs/tsleeps/wakeups when a pipe
529 		 * is used for synchronous communications with small packets,
530 		 * and 8000 or so (8uS) will pipeline large buffer xfers
531 		 * between cpus over a pipe.
532 		 *
533 		 * For synchronous communications a hit means doing a
534 		 * full Awrite-Bread-Bwrite-Aread cycle in less then 2uS,
535 		 * where as miss requiring a tsleep/wakeup sequence
536 		 * will take 7uS or more.
537 		 */
538 		if (rpipe->pipe_buffer.windex != rpipe->pipe_buffer.rindex)
539 			continue;
540 
541 #ifdef _RDTSC_SUPPORTED_
542 		if (pipe_delay) {
543 			int64_t tsc_target;
544 			int good = 0;
545 
546 			tsc_target = tsc_get_target(pipe_delay);
547 			while (tsc_test_target(tsc_target) == 0) {
548 				if (rpipe->pipe_buffer.windex !=
549 				    rpipe->pipe_buffer.rindex) {
550 					good = 1;
551 					break;
552 				}
553 			}
554 			if (good)
555 				continue;
556 		}
557 #endif
558 
559 		/*
560 		 * Detect EOF condition, do not set error.
561 		 */
562 		if (rpipe->pipe_state & PIPE_REOF)
563 			break;
564 
565 		/*
566 		 * Break if some data was read, or if this was a non-blocking
567 		 * read.
568 		 */
569 		if (nread > 0)
570 			break;
571 
572 		if (nbio) {
573 			error = EAGAIN;
574 			break;
575 		}
576 
577 		/*
578 		 * Last chance, interlock with WANTR.
579 		 */
580 		lwkt_gettoken(&rpipe->pipe_wlock);
581 		size = rpipe->pipe_buffer.windex - rpipe->pipe_buffer.rindex;
582 		if (size) {
583 			lwkt_reltoken(&rpipe->pipe_wlock);
584 			continue;
585 		}
586 
587 		/*
588 		 * Retest EOF - acquiring a new token can temporarily release
589 		 * tokens already held.
590 		 */
591 		if (rpipe->pipe_state & PIPE_REOF) {
592 			lwkt_reltoken(&rpipe->pipe_wlock);
593 			break;
594 		}
595 
596 		/*
597 		 * If there is no more to read in the pipe, reset its
598 		 * pointers to the beginning.  This improves cache hit
599 		 * stats.
600 		 *
601 		 * We need both locks to modify both pointers, and there
602 		 * must also not be a write in progress or the uiomove()
603 		 * in the write might block and temporarily release
604 		 * its wlock, then reacquire and update windex.  We are
605 		 * only serialized against reads, not writes.
606 		 *
607 		 * XXX should we even bother resetting the indices?  It
608 		 *     might actually be more cache efficient not to.
609 		 */
610 		if (rpipe->pipe_buffer.rindex == rpipe->pipe_buffer.windex &&
611 		    rpipe->pipe_wip == 0) {
612 			rpipe->pipe_buffer.rindex = 0;
613 			rpipe->pipe_buffer.windex = 0;
614 		}
615 
616 		/*
617 		 * Wait for more data.
618 		 *
619 		 * Pipe_state can only be set if both the rlock and wlock
620 		 * are held.
621 		 */
622 		rpipe->pipe_state |= PIPE_WANTR;
623 		tsleep_interlock(rpipe, PCATCH);
624 		lwkt_reltoken(&rpipe->pipe_wlock);
625 		error = tsleep(rpipe, PCATCH | PINTERLOCKED, "piperd", 0);
626 		++pipe_rblocked_count;
627 		if (error)
628 			break;
629 	}
630 	pipe_end_uio(rpipe, &rpipe->pipe_rip);
631 
632 	/*
633 	 * Uptime last access time
634 	 */
635 	if (error == 0 && nread)
636 		vfs_timestamp(&rpipe->pipe_atime);
637 
638 	/*
639 	 * If we drained the FIFO more then half way then handle
640 	 * write blocking hysteresis.
641 	 *
642 	 * Note that PIPE_WANTW cannot be set by the writer without
643 	 * it holding both rlock and wlock, so we can test it
644 	 * while holding just rlock.
645 	 */
646 	if (notify_writer) {
647 		/*
648 		 * Synchronous blocking is done on the pipe involved
649 		 */
650 		if (rpipe->pipe_state & PIPE_WANTW) {
651 			lwkt_gettoken(&rpipe->pipe_wlock);
652 			if (rpipe->pipe_state & PIPE_WANTW) {
653 				rpipe->pipe_state &= ~PIPE_WANTW;
654 				lwkt_reltoken(&rpipe->pipe_wlock);
655 				wakeup(rpipe);
656 			} else {
657 				lwkt_reltoken(&rpipe->pipe_wlock);
658 			}
659 		}
660 
661 		/*
662 		 * But we may also have to deal with a kqueue which is
663 		 * stored on the same pipe as its descriptor, so a
664 		 * EVFILT_WRITE event waiting for our side to drain will
665 		 * be on the other side.
666 		 */
667 		lwkt_gettoken(&wpipe->pipe_wlock);
668 		pipewakeup(wpipe, 0);
669 		lwkt_reltoken(&wpipe->pipe_wlock);
670 	}
671 	/*size = rpipe->pipe_buffer.windex - rpipe->pipe_buffer.rindex;*/
672 	lwkt_reltoken(&rpipe->pipe_rlock);
673 
674 	return (error);
675 }
676 
677 static int
678 pipe_write(struct file *fp, struct uio *uio, struct ucred *cred, int fflags)
679 {
680 	int error;
681 	int orig_resid;
682 	int nbio;
683 	struct pipe *wpipe;
684 	struct pipe *rpipe;
685 	u_int windex;
686 	u_int space;
687 	u_int wcount;
688 	int bigwrite;
689 	int bigcount;
690 
691 	/*
692 	 * Writes go to the peer.  The peer will always exist.
693 	 */
694 	rpipe = (struct pipe *) fp->f_data;
695 	wpipe = rpipe->pipe_peer;
696 	lwkt_gettoken(&wpipe->pipe_wlock);
697 	if (wpipe->pipe_state & PIPE_WEOF) {
698 		lwkt_reltoken(&wpipe->pipe_wlock);
699 		return (EPIPE);
700 	}
701 
702 	/*
703 	 * Degenerate case (EPIPE takes prec)
704 	 */
705 	if (uio->uio_resid == 0) {
706 		lwkt_reltoken(&wpipe->pipe_wlock);
707 		return(0);
708 	}
709 
710 	/*
711 	 * Writes are serialized (start_uio must be called with wlock)
712 	 */
713 	error = pipe_start_uio(wpipe, &wpipe->pipe_wip);
714 	if (error) {
715 		lwkt_reltoken(&wpipe->pipe_wlock);
716 		return (error);
717 	}
718 
719 	if (fflags & O_FBLOCKING)
720 		nbio = 0;
721 	else if (fflags & O_FNONBLOCKING)
722 		nbio = 1;
723 	else if (fp->f_flag & O_NONBLOCK)
724 		nbio = 1;
725 	else
726 		nbio = 0;
727 
728 	/*
729 	 * If it is advantageous to resize the pipe buffer, do
730 	 * so.  We are write-serialized so we can block safely.
731 	 */
732 	if ((wpipe->pipe_buffer.size <= PIPE_SIZE) &&
733 	    (pipe_nbig < pipe_maxbig) &&
734 	    wpipe->pipe_wantwcnt > 4 &&
735 	    (wpipe->pipe_buffer.rindex == wpipe->pipe_buffer.windex)) {
736 		/*
737 		 * Recheck after lock.
738 		 */
739 		lwkt_gettoken(&wpipe->pipe_rlock);
740 		if ((wpipe->pipe_buffer.size <= PIPE_SIZE) &&
741 		    (pipe_nbig < pipe_maxbig) &&
742 		    (wpipe->pipe_buffer.rindex == wpipe->pipe_buffer.windex)) {
743 			atomic_add_int(&pipe_nbig, 1);
744 			if (pipespace(wpipe, BIG_PIPE_SIZE) == 0)
745 				++pipe_bigcount;
746 			else
747 				atomic_subtract_int(&pipe_nbig, 1);
748 		}
749 		lwkt_reltoken(&wpipe->pipe_rlock);
750 	}
751 
752 	orig_resid = uio->uio_resid;
753 	wcount = 0;
754 
755 	bigwrite = (uio->uio_resid > 10 * 1024 * 1024);
756 	bigcount = 10;
757 
758 	while (uio->uio_resid) {
759 		if (wpipe->pipe_state & PIPE_WEOF) {
760 			error = EPIPE;
761 			break;
762 		}
763 
764 		/*
765 		 * Don't hog the cpu.
766 		 */
767 		if (bigwrite && --bigcount == 0) {
768 			lwkt_user_yield();
769 			bigcount = 10;
770 			if (CURSIG(curthread->td_lwp)) {
771 				error = EINTR;
772 				break;
773 			}
774 		}
775 
776 		windex = wpipe->pipe_buffer.windex &
777 			 (wpipe->pipe_buffer.size - 1);
778 		space = wpipe->pipe_buffer.size -
779 			(wpipe->pipe_buffer.windex - wpipe->pipe_buffer.rindex);
780 		cpu_lfence();
781 
782 		/* Writes of size <= PIPE_BUF must be atomic. */
783 		if ((space < uio->uio_resid) && (orig_resid <= PIPE_BUF))
784 			space = 0;
785 
786 		/*
787 		 * Write to fill, read size handles write hysteresis.  Also
788 		 * additional restrictions can cause select-based non-blocking
789 		 * writes to spin.
790 		 */
791 		if (space > 0) {
792 			u_int segsize;
793 
794 			/*
795 			 * Transfer size is minimum of uio transfer
796 			 * and free space in pipe buffer.
797 			 *
798 			 * Limit each uiocopy to no more then PIPE_SIZE
799 			 * so we can keep the gravy train going on a
800 			 * SMP box.  This doubles the performance for
801 			 * write sizes > 16K.  Otherwise large writes
802 			 * wind up doing an inefficient synchronous
803 			 * ping-pong.
804 			 */
805 			space = szmin(space, uio->uio_resid);
806 			if (space > PIPE_SIZE)
807 				space = PIPE_SIZE;
808 
809 			/*
810 			 * First segment to transfer is minimum of
811 			 * transfer size and contiguous space in
812 			 * pipe buffer.  If first segment to transfer
813 			 * is less than the transfer size, we've got
814 			 * a wraparound in the buffer.
815 			 */
816 			segsize = wpipe->pipe_buffer.size - windex;
817 			if (segsize > space)
818 				segsize = space;
819 
820 			/*
821 			 * If this is the first loop and the reader is
822 			 * blocked, do a preemptive wakeup of the reader.
823 			 *
824 			 * On SMP the IPI latency plus the wlock interlock
825 			 * on the reader side is the fastest way to get the
826 			 * reader going.  (The scheduler will hard loop on
827 			 * lock tokens).
828 			 *
829 			 * NOTE: We can't clear WANTR here without acquiring
830 			 * the rlock, which we don't want to do here!
831 			 */
832 			if ((wpipe->pipe_state & PIPE_WANTR))
833 				wakeup(wpipe);
834 
835 			/*
836 			 * Transfer segment, which may include a wrap-around.
837 			 * Update windex to account for both all in one go
838 			 * so the reader can read() the data atomically.
839 			 */
840 			error = uiomove(&wpipe->pipe_buffer.buffer[windex],
841 					segsize, uio);
842 			if (error == 0 && segsize < space) {
843 				segsize = space - segsize;
844 				error = uiomove(&wpipe->pipe_buffer.buffer[0],
845 						segsize, uio);
846 			}
847 			if (error)
848 				break;
849 			cpu_mfence();
850 			wpipe->pipe_buffer.windex += space;
851 			wcount += space;
852 			continue;
853 		}
854 
855 		/*
856 		 * We need both the rlock and the wlock to interlock against
857 		 * the EOF, WANTW, and size checks, and to modify pipe_state.
858 		 *
859 		 * These are token locks so we do not have to worry about
860 		 * deadlocks.
861 		 */
862 		lwkt_gettoken(&wpipe->pipe_rlock);
863 
864 		/*
865 		 * If the "read-side" has been blocked, wake it up now
866 		 * and yield to let it drain synchronously rather
867 		 * then block.
868 		 */
869 		if (wpipe->pipe_state & PIPE_WANTR) {
870 			wpipe->pipe_state &= ~PIPE_WANTR;
871 			wakeup(wpipe);
872 		}
873 
874 		/*
875 		 * don't block on non-blocking I/O
876 		 */
877 		if (nbio) {
878 			lwkt_reltoken(&wpipe->pipe_rlock);
879 			error = EAGAIN;
880 			break;
881 		}
882 
883 		/*
884 		 * re-test whether we have to block in the writer after
885 		 * acquiring both locks, in case the reader opened up
886 		 * some space.
887 		 */
888 		space = wpipe->pipe_buffer.size -
889 			(wpipe->pipe_buffer.windex - wpipe->pipe_buffer.rindex);
890 		cpu_lfence();
891 		if ((space < uio->uio_resid) && (orig_resid <= PIPE_BUF))
892 			space = 0;
893 
894 		/*
895 		 * Retest EOF - acquiring a new token can temporarily release
896 		 * tokens already held.
897 		 */
898 		if (wpipe->pipe_state & PIPE_WEOF) {
899 			lwkt_reltoken(&wpipe->pipe_rlock);
900 			error = EPIPE;
901 			break;
902 		}
903 
904 		/*
905 		 * We have no more space and have something to offer,
906 		 * wake up select/poll/kq.
907 		 */
908 		if (space == 0) {
909 			wpipe->pipe_state |= PIPE_WANTW;
910 			++wpipe->pipe_wantwcnt;
911 			pipewakeup(wpipe, 1);
912 			if (wpipe->pipe_state & PIPE_WANTW)
913 				error = tsleep(wpipe, PCATCH, "pipewr", 0);
914 			++pipe_wblocked_count;
915 		}
916 		lwkt_reltoken(&wpipe->pipe_rlock);
917 
918 		/*
919 		 * Break out if we errored or the read side wants us to go
920 		 * away.
921 		 */
922 		if (error)
923 			break;
924 		if (wpipe->pipe_state & PIPE_WEOF) {
925 			error = EPIPE;
926 			break;
927 		}
928 	}
929 	pipe_end_uio(wpipe, &wpipe->pipe_wip);
930 
931 	/*
932 	 * If we have put any characters in the buffer, we wake up
933 	 * the reader.
934 	 *
935 	 * Both rlock and wlock are required to be able to modify pipe_state.
936 	 */
937 	if (wpipe->pipe_buffer.windex != wpipe->pipe_buffer.rindex) {
938 		if (wpipe->pipe_state & PIPE_WANTR) {
939 			lwkt_gettoken(&wpipe->pipe_rlock);
940 			if (wpipe->pipe_state & PIPE_WANTR) {
941 				wpipe->pipe_state &= ~PIPE_WANTR;
942 				lwkt_reltoken(&wpipe->pipe_rlock);
943 				wakeup(wpipe);
944 			} else {
945 				lwkt_reltoken(&wpipe->pipe_rlock);
946 			}
947 		}
948 		lwkt_gettoken(&wpipe->pipe_rlock);
949 		pipewakeup(wpipe, 1);
950 		lwkt_reltoken(&wpipe->pipe_rlock);
951 	}
952 
953 	/*
954 	 * Don't return EPIPE if I/O was successful
955 	 */
956 	if ((wpipe->pipe_buffer.rindex == wpipe->pipe_buffer.windex) &&
957 	    (uio->uio_resid == 0) &&
958 	    (error == EPIPE)) {
959 		error = 0;
960 	}
961 
962 	if (error == 0)
963 		vfs_timestamp(&wpipe->pipe_mtime);
964 
965 	/*
966 	 * We have something to offer,
967 	 * wake up select/poll/kq.
968 	 */
969 	/*space = wpipe->pipe_buffer.windex - wpipe->pipe_buffer.rindex;*/
970 	lwkt_reltoken(&wpipe->pipe_wlock);
971 	return (error);
972 }
973 
974 /*
975  * we implement a very minimal set of ioctls for compatibility with sockets.
976  */
977 static int
978 pipe_ioctl(struct file *fp, u_long cmd, caddr_t data,
979 	   struct ucred *cred, struct sysmsg *msg)
980 {
981 	struct pipe *mpipe;
982 	int error;
983 
984 	mpipe = (struct pipe *)fp->f_data;
985 
986 	lwkt_gettoken(&mpipe->pipe_rlock);
987 	lwkt_gettoken(&mpipe->pipe_wlock);
988 
989 	switch (cmd) {
990 	case FIOASYNC:
991 		if (*(int *)data) {
992 			mpipe->pipe_state |= PIPE_ASYNC;
993 		} else {
994 			mpipe->pipe_state &= ~PIPE_ASYNC;
995 		}
996 		error = 0;
997 		break;
998 	case FIONREAD:
999 		*(int *)data = mpipe->pipe_buffer.windex -
1000 				mpipe->pipe_buffer.rindex;
1001 		error = 0;
1002 		break;
1003 	case FIOSETOWN:
1004 		error = fsetown(*(int *)data, &mpipe->pipe_sigio);
1005 		break;
1006 	case FIOGETOWN:
1007 		*(int *)data = fgetown(&mpipe->pipe_sigio);
1008 		error = 0;
1009 		break;
1010 	case TIOCSPGRP:
1011 		/* This is deprecated, FIOSETOWN should be used instead. */
1012 		error = fsetown(-(*(int *)data), &mpipe->pipe_sigio);
1013 		break;
1014 
1015 	case TIOCGPGRP:
1016 		/* This is deprecated, FIOGETOWN should be used instead. */
1017 		*(int *)data = -fgetown(&mpipe->pipe_sigio);
1018 		error = 0;
1019 		break;
1020 	default:
1021 		error = ENOTTY;
1022 		break;
1023 	}
1024 	lwkt_reltoken(&mpipe->pipe_wlock);
1025 	lwkt_reltoken(&mpipe->pipe_rlock);
1026 
1027 	return (error);
1028 }
1029 
1030 /*
1031  * MPSAFE
1032  */
1033 static int
1034 pipe_stat(struct file *fp, struct stat *ub, struct ucred *cred)
1035 {
1036 	struct pipe *pipe;
1037 
1038 	pipe = (struct pipe *)fp->f_data;
1039 
1040 	bzero((caddr_t)ub, sizeof(*ub));
1041 	ub->st_mode = S_IFIFO;
1042 	ub->st_blksize = pipe->pipe_buffer.size;
1043 	ub->st_size = pipe->pipe_buffer.windex - pipe->pipe_buffer.rindex;
1044 	ub->st_blocks = (ub->st_size + ub->st_blksize - 1) / ub->st_blksize;
1045 	ub->st_atimespec = pipe->pipe_atime;
1046 	ub->st_mtimespec = pipe->pipe_mtime;
1047 	ub->st_ctimespec = pipe->pipe_ctime;
1048 	/*
1049 	 * Left as 0: st_dev, st_ino, st_nlink, st_uid, st_gid, st_rdev,
1050 	 * st_flags, st_gen.
1051 	 * XXX (st_dev, st_ino) should be unique.
1052 	 */
1053 	return (0);
1054 }
1055 
1056 static int
1057 pipe_close(struct file *fp)
1058 {
1059 	struct pipe *cpipe;
1060 
1061 	cpipe = (struct pipe *)fp->f_data;
1062 	fp->f_ops = &badfileops;
1063 	fp->f_data = NULL;
1064 	funsetown(&cpipe->pipe_sigio);
1065 	pipeclose(cpipe);
1066 	return (0);
1067 }
1068 
1069 /*
1070  * Shutdown one or both directions of a full-duplex pipe.
1071  */
1072 static int
1073 pipe_shutdown(struct file *fp, int how)
1074 {
1075 	struct pipe *rpipe;
1076 	struct pipe *wpipe;
1077 	int error = EPIPE;
1078 
1079 	rpipe = (struct pipe *)fp->f_data;
1080 	wpipe = rpipe->pipe_peer;
1081 
1082 	/*
1083 	 * We modify pipe_state on both pipes, which means we need
1084 	 * all four tokens!
1085 	 */
1086 	lwkt_gettoken(&rpipe->pipe_rlock);
1087 	lwkt_gettoken(&rpipe->pipe_wlock);
1088 	lwkt_gettoken(&wpipe->pipe_rlock);
1089 	lwkt_gettoken(&wpipe->pipe_wlock);
1090 
1091 	switch(how) {
1092 	case SHUT_RDWR:
1093 	case SHUT_RD:
1094 		rpipe->pipe_state |= PIPE_REOF;		/* my reads */
1095 		rpipe->pipe_state |= PIPE_WEOF;		/* peer writes */
1096 		if (rpipe->pipe_state & PIPE_WANTR) {
1097 			rpipe->pipe_state &= ~PIPE_WANTR;
1098 			wakeup(rpipe);
1099 		}
1100 		if (rpipe->pipe_state & PIPE_WANTW) {
1101 			rpipe->pipe_state &= ~PIPE_WANTW;
1102 			wakeup(rpipe);
1103 		}
1104 		error = 0;
1105 		if (how == SHUT_RD)
1106 			break;
1107 		/* fall through */
1108 	case SHUT_WR:
1109 		wpipe->pipe_state |= PIPE_REOF;		/* peer reads */
1110 		wpipe->pipe_state |= PIPE_WEOF;		/* my writes */
1111 		if (wpipe->pipe_state & PIPE_WANTR) {
1112 			wpipe->pipe_state &= ~PIPE_WANTR;
1113 			wakeup(wpipe);
1114 		}
1115 		if (wpipe->pipe_state & PIPE_WANTW) {
1116 			wpipe->pipe_state &= ~PIPE_WANTW;
1117 			wakeup(wpipe);
1118 		}
1119 		error = 0;
1120 		break;
1121 	}
1122 	pipewakeup(rpipe, 1);
1123 	pipewakeup(wpipe, 1);
1124 
1125 	lwkt_reltoken(&wpipe->pipe_wlock);
1126 	lwkt_reltoken(&wpipe->pipe_rlock);
1127 	lwkt_reltoken(&rpipe->pipe_wlock);
1128 	lwkt_reltoken(&rpipe->pipe_rlock);
1129 
1130 	return (error);
1131 }
1132 
1133 static void
1134 pipe_free_kmem(struct pipe *cpipe)
1135 {
1136 	if (cpipe->pipe_buffer.buffer != NULL) {
1137 		if (cpipe->pipe_buffer.size > PIPE_SIZE)
1138 			atomic_subtract_int(&pipe_nbig, 1);
1139 		kmem_free(&kernel_map,
1140 			(vm_offset_t)cpipe->pipe_buffer.buffer,
1141 			cpipe->pipe_buffer.size);
1142 		cpipe->pipe_buffer.buffer = NULL;
1143 		cpipe->pipe_buffer.object = NULL;
1144 	}
1145 }
1146 
1147 /*
1148  * Close the pipe.  The slock must be held to interlock against simultanious
1149  * closes.  The rlock and wlock must be held to adjust the pipe_state.
1150  */
1151 static void
1152 pipeclose(struct pipe *cpipe)
1153 {
1154 	globaldata_t gd;
1155 	struct pipe *ppipe;
1156 
1157 	if (cpipe == NULL)
1158 		return;
1159 
1160 	/*
1161 	 * The slock may not have been allocated yet (close during
1162 	 * initialization)
1163 	 *
1164 	 * We need both the read and write tokens to modify pipe_state.
1165 	 */
1166 	if (cpipe->pipe_slock)
1167 		lockmgr(cpipe->pipe_slock, LK_EXCLUSIVE);
1168 	lwkt_gettoken(&cpipe->pipe_rlock);
1169 	lwkt_gettoken(&cpipe->pipe_wlock);
1170 
1171 	/*
1172 	 * Set our state, wakeup anyone waiting in select/poll/kq, and
1173 	 * wakeup anyone blocked on our pipe.
1174 	 */
1175 	cpipe->pipe_state |= PIPE_CLOSED | PIPE_REOF | PIPE_WEOF;
1176 	pipewakeup(cpipe, 1);
1177 	if (cpipe->pipe_state & (PIPE_WANTR | PIPE_WANTW)) {
1178 		cpipe->pipe_state &= ~(PIPE_WANTR | PIPE_WANTW);
1179 		wakeup(cpipe);
1180 	}
1181 
1182 	/*
1183 	 * Disconnect from peer.
1184 	 */
1185 	if ((ppipe = cpipe->pipe_peer) != NULL) {
1186 		lwkt_gettoken(&ppipe->pipe_rlock);
1187 		lwkt_gettoken(&ppipe->pipe_wlock);
1188 		ppipe->pipe_state |= PIPE_REOF | PIPE_WEOF;
1189 		pipewakeup(ppipe, 1);
1190 		if (ppipe->pipe_state & (PIPE_WANTR | PIPE_WANTW)) {
1191 			ppipe->pipe_state &= ~(PIPE_WANTR | PIPE_WANTW);
1192 			wakeup(ppipe);
1193 		}
1194 		if (SLIST_FIRST(&ppipe->pipe_kq.ki_note))
1195 			KNOTE(&ppipe->pipe_kq.ki_note, 0);
1196 		lwkt_reltoken(&ppipe->pipe_wlock);
1197 		lwkt_reltoken(&ppipe->pipe_rlock);
1198 	}
1199 
1200 	/*
1201 	 * If the peer is also closed we can free resources for both
1202 	 * sides, otherwise we leave our side intact to deal with any
1203 	 * races (since we only have the slock).
1204 	 */
1205 	if (ppipe && (ppipe->pipe_state & PIPE_CLOSED)) {
1206 		cpipe->pipe_peer = NULL;
1207 		ppipe->pipe_peer = NULL;
1208 		ppipe->pipe_slock = NULL;	/* we will free the slock */
1209 		pipeclose(ppipe);
1210 		ppipe = NULL;
1211 	}
1212 
1213 	lwkt_reltoken(&cpipe->pipe_wlock);
1214 	lwkt_reltoken(&cpipe->pipe_rlock);
1215 	if (cpipe->pipe_slock)
1216 		lockmgr(cpipe->pipe_slock, LK_RELEASE);
1217 
1218 	/*
1219 	 * If we disassociated from our peer we can free resources
1220 	 */
1221 	if (ppipe == NULL) {
1222 		gd = mycpu;
1223 		if (cpipe->pipe_slock) {
1224 			kfree(cpipe->pipe_slock, M_PIPE);
1225 			cpipe->pipe_slock = NULL;
1226 		}
1227 		if (gd->gd_pipeqcount >= pipe_maxcache ||
1228 		    cpipe->pipe_buffer.size != PIPE_SIZE
1229 		) {
1230 			pipe_free_kmem(cpipe);
1231 			kfree(cpipe, M_PIPE);
1232 		} else {
1233 			cpipe->pipe_state = 0;
1234 			cpipe->pipe_peer = gd->gd_pipeq;
1235 			gd->gd_pipeq = cpipe;
1236 			++gd->gd_pipeqcount;
1237 		}
1238 	}
1239 }
1240 
1241 static int
1242 pipe_kqfilter(struct file *fp, struct knote *kn)
1243 {
1244 	struct pipe *cpipe;
1245 
1246 	cpipe = (struct pipe *)kn->kn_fp->f_data;
1247 
1248 	switch (kn->kn_filter) {
1249 	case EVFILT_READ:
1250 		kn->kn_fop = &pipe_rfiltops;
1251 		break;
1252 	case EVFILT_WRITE:
1253 		kn->kn_fop = &pipe_wfiltops;
1254 		if (cpipe->pipe_peer == NULL) {
1255 			/* other end of pipe has been closed */
1256 			return (EPIPE);
1257 		}
1258 		break;
1259 	default:
1260 		return (EOPNOTSUPP);
1261 	}
1262 	kn->kn_hook = (caddr_t)cpipe;
1263 
1264 	knote_insert(&cpipe->pipe_kq.ki_note, kn);
1265 
1266 	return (0);
1267 }
1268 
1269 static void
1270 filt_pipedetach(struct knote *kn)
1271 {
1272 	struct pipe *cpipe = (struct pipe *)kn->kn_hook;
1273 
1274 	knote_remove(&cpipe->pipe_kq.ki_note, kn);
1275 }
1276 
1277 /*ARGSUSED*/
1278 static int
1279 filt_piperead(struct knote *kn, long hint)
1280 {
1281 	struct pipe *rpipe = (struct pipe *)kn->kn_fp->f_data;
1282 	int ready = 0;
1283 
1284 	lwkt_gettoken(&rpipe->pipe_rlock);
1285 	lwkt_gettoken(&rpipe->pipe_wlock);
1286 
1287 	kn->kn_data = rpipe->pipe_buffer.windex - rpipe->pipe_buffer.rindex;
1288 
1289 	if (rpipe->pipe_state & PIPE_REOF) {
1290 		/*
1291 		 * Only set NODATA if all data has been exhausted
1292 		 */
1293 		if (kn->kn_data == 0)
1294 			kn->kn_flags |= EV_NODATA;
1295 		kn->kn_flags |= EV_EOF;
1296 		ready = 1;
1297 	}
1298 
1299 	lwkt_reltoken(&rpipe->pipe_wlock);
1300 	lwkt_reltoken(&rpipe->pipe_rlock);
1301 
1302 	if (!ready)
1303 		ready = kn->kn_data > 0;
1304 
1305 	return (ready);
1306 }
1307 
1308 /*ARGSUSED*/
1309 static int
1310 filt_pipewrite(struct knote *kn, long hint)
1311 {
1312 	struct pipe *rpipe = (struct pipe *)kn->kn_fp->f_data;
1313 	struct pipe *wpipe = rpipe->pipe_peer;
1314 	int ready = 0;
1315 
1316 	kn->kn_data = 0;
1317 	if (wpipe == NULL) {
1318 		kn->kn_flags |= (EV_EOF | EV_NODATA);
1319 		return (1);
1320 	}
1321 
1322 	lwkt_gettoken(&wpipe->pipe_rlock);
1323 	lwkt_gettoken(&wpipe->pipe_wlock);
1324 
1325 	if (wpipe->pipe_state & PIPE_WEOF) {
1326 		kn->kn_flags |= (EV_EOF | EV_NODATA);
1327 		ready = 1;
1328 	}
1329 
1330 	if (!ready)
1331 		kn->kn_data = wpipe->pipe_buffer.size -
1332 			      (wpipe->pipe_buffer.windex -
1333 			       wpipe->pipe_buffer.rindex);
1334 
1335 	lwkt_reltoken(&wpipe->pipe_wlock);
1336 	lwkt_reltoken(&wpipe->pipe_rlock);
1337 
1338 	if (!ready)
1339 		ready = kn->kn_data >= PIPE_BUF;
1340 
1341 	return (ready);
1342 }
1343