/*-
* Copyright (c) 2003, 2007, 2008, 2009, 2023 The NetBSD Foundation, Inc.
* All rights reserved.
*
* This code is derived from software contributed to The NetBSD Foundation
* by Paul Kranenburg, and by Andrew Doran.
*
* Redistribution and use in source and binary forms, with or without
* modification, are permitted provided that the following conditions
* are met:
* 1. Redistributions of source code must retain the above copyright
* notice, this list of conditions and the following disclaimer.
* 2. Redistributions in binary form must reproduce the above copyright
* notice, this list of conditions and the following disclaimer in the
* documentation and/or other materials provided with the distribution.
*
* THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
* ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
* TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
* PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
* BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
* CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
* SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
* INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
* CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
* ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
* POSSIBILITY OF SUCH DAMAGE.
*/
/*
* Copyright (c) 1996 John S. Dyson
* All rights reserved.
*
* Redistribution and use in source and binary forms, with or without
* modification, are permitted provided that the following conditions
* are met:
* 1. Redistributions of source code must retain the above copyright
* notice immediately at the beginning of the file, without modification,
* this list of conditions, and the following disclaimer.
* 2. Redistributions in binary form must reproduce the above copyright
* notice, this list of conditions and the following disclaimer in the
* documentation and/or other materials provided with the distribution.
* 3. Absolutely no warranty of function or purpose is made by the author
* John S. Dyson.
* 4. Modifications may be freely made to this file if the above conditions
* are met.
*/
/*
* This file contains a high-performance replacement for the socket-based
* pipes scheme originally used. It does not support all features of
* sockets, but does do everything that pipes normally do.
*/
/*
* Default pipe buffer size(s), this can be kind-of large now because pipe
* space is pageable. The pipe code will try to maintain locality of
* reference for performance reasons, so small amounts of outstanding I/O
* will not wipe the cache.
*/
#define MINPIPESIZE (PIPE_SIZE / 3)
#define MAXPIPESIZE (2 * PIPE_SIZE / 3)
/*
* Limit the number of "big" pipes
*/
#define LIMITBIGPIPES 32
static u_int maxbigpipes __read_mostly = LIMITBIGPIPES;
static u_int nbigpipe = 0;
/*
* Amount of KVA consumed by pipe buffers.
*/
static u_int amountpipekva = 0;
/*
* The pipe system call for the DTYPE_PIPE type of pipes
*/
int
pipe1(struct lwp *l, int *fildes, int flags)
{
struct pipe *rpipe, *wpipe;
struct timespec nt;
file_t *rf, *wf;
int fd, error;
proc_t *p;
/*
* Allocate kva for pipe circular buffer, the space is pageable
* This routine will 'realloc' the size of a pipe safely, if it fails
* it will retain the old buffer.
* If it fails it will return ENOMEM.
*/
static int
pipespace(struct pipe *pipe, int size)
{
void *buffer;
/*
* Initialize and allocate VM and memory for pipe.
*/
static int
pipe_create(struct pipe **pipep, pool_cache_t cache, struct timespec *nt)
{
struct pipe *pipe;
int error;
/*
* Lock a pipe for I/O, blocking other access
* Called with pipe spin lock held.
*/
static int
pipelock(struct pipe *pipe, bool catch_p)
{
int error;
KASSERT(mutex_owned(pipe->pipe_lock));
while (pipe->pipe_state & PIPE_LOCKFL) {
if (catch_p) {
error = cv_wait_sig(&pipe->pipe_lkcv, pipe->pipe_lock);
if (error != 0) {
return error;
}
} else
cv_wait(&pipe->pipe_lkcv, pipe->pipe_lock);
}
/*
* Select/poll wakeup. This also sends SIGIO to peer connected to
* 'sigpipe' side of pipe.
*/
static void
pipeselwakeup(struct pipe *selp, struct pipe *sigp, int code)
{
int band;
switch (code) {
case POLL_IN:
band = POLLIN|POLLRDNORM;
break;
case POLL_OUT:
band = POLLOUT|POLLWRNORM;
break;
case POLL_HUP:
band = POLLHUP;
break;
case POLL_ERR:
band = POLLERR;
break;
default:
band = 0;
#ifdef DIAGNOSTIC
printf("bad siginfo code %d in pipe notification.\n", code);
#endif
break;
}
/*
* Try to avoid locking the pipe if we have nothing to do.
*
* There are programs which share one pipe amongst multiple processes
* and perform non-blocking reads in parallel, even if the pipe is
* empty. This in particular is the case with BSD make, which when
* spawned with a high -j number can find itself with over half of the
* calls failing to find anything.
*/
if ((fp->f_flag & FNONBLOCK) != 0) {
if (__predict_false(uio->uio_resid == 0))
return (0);
if (atomic_load_relaxed(&bp->cnt) == 0 &&
(atomic_load_relaxed(&rpipe->pipe_state) & PIPE_EOF) == 0)
return (EAGAIN);
}
bp->out += size;
if (bp->out >= bp->size)
bp->out = 0;
bp->cnt -= size;
/*
* If there is no more to read in the pipe, reset
* its pointers to the beginning. This improves
* cache hit stats.
*/
if (bp->cnt == 0) {
bp->in = 0;
bp->out = 0;
}
nread += size;
continue;
}
/*
* Break if some data was read.
*/
if (nread > 0)
break;
/*
* Detect EOF condition.
* Read returns 0 on EOF, no need to set error.
*/
if (rpipe->pipe_state & PIPE_EOF)
break;
/*
* Don't block on non-blocking I/O.
*/
if (fp->f_flag & FNONBLOCK) {
error = EAGAIN;
break;
}
/*
* Unlock the pipe buffer for our remaining processing.
* We will either break out with an error or we will
* sleep and relock to loop.
*/
pipeunlock(rpipe);
#if 1 /* XXX (dsl) I'm sure these aren't needed here ... */
/*
* We want to read more, wake up select/poll.
*/
pipeselwakeup(rpipe, rpipe->pipe_peer, POLL_OUT);
/*
* If the "write-side" is blocked, wake it up now.
*/
cv_broadcast(&rpipe->pipe_wcv);
#endif
/* Now wait until the pipe is filled */
error = cv_wait_sig(&rpipe->pipe_rcv, lock);
if (error != 0)
goto unlocked_error;
wakeup_state = rpipe->pipe_state;
goto again;
}
if (error == 0)
getnanotime(&rpipe->pipe_atime);
pipeunlock(rpipe);
unlocked_error:
--rpipe->pipe_busy;
if (rpipe->pipe_busy == 0) {
rpipe->pipe_state &= ~PIPE_RESTART;
cv_broadcast(&rpipe->pipe_draincv);
}
if (bp->cnt < MINPIPESIZE) {
cv_broadcast(&rpipe->pipe_wcv);
}
/*
* If anything was read off the buffer, signal to the writer it's
* possible to write more data. Also send signal if we are here for the
* first time after last write.
*/
if ((bp->size - bp->cnt) >= PIPE_BUF
&& (ocnt != bp->cnt || (rpipe->pipe_state & PIPE_SIGNALR))) {
pipeselwakeup(rpipe, rpipe->pipe_peer, POLL_OUT);
rpipe->pipe_state &= ~PIPE_SIGNALR;
}
mutex_exit(lock);
return (error);
}
static int
pipe_write(file_t *fp, off_t *offset, struct uio *uio, kauth_cred_t cred,
int flags)
{
struct pipe *wpipe, *rpipe;
struct pipebuf *bp;
kmutex_t *lock;
int error;
unsigned int wakeup_state = 0;
/* We want to write to our peer */
rpipe = fp->f_pipe;
lock = rpipe->pipe_lock;
error = 0;
mutex_enter(lock);
wpipe = rpipe->pipe_peer;
/*
* Detect loss of pipe read side, issue SIGPIPE if lost.
*/
if (wpipe == NULL || (wpipe->pipe_state & PIPE_EOF) != 0) {
mutex_exit(lock);
return EPIPE;
}
++wpipe->pipe_busy;
/* Acquire the long-term pipe lock */
if ((error = pipelock(wpipe, true)) != 0) {
--wpipe->pipe_busy;
if (wpipe->pipe_busy == 0) {
wpipe->pipe_state &= ~PIPE_RESTART;
cv_broadcast(&wpipe->pipe_draincv);
}
mutex_exit(lock);
return (error);
}
bp = &wpipe->pipe_buffer;
/*
* If it is advantageous to resize the pipe buffer, do so.
*/
if ((uio->uio_resid > PIPE_SIZE) &&
(nbigpipe < maxbigpipes) &&
(bp->size <= PIPE_SIZE) && (bp->cnt == 0)) {
if (pipespace(wpipe, BIG_PIPE_SIZE) == 0)
atomic_inc_uint(&nbigpipe);
}
while (uio->uio_resid) {
size_t space;
space = bp->size - bp->cnt;
/* Writes of size <= PIPE_BUF must be atomic. */
if ((space < uio->uio_resid) && (uio->uio_resid <= PIPE_BUF))
space = 0;
if (space > 0) {
int size; /* Transfer size */
int segsize; /* first segment to transfer */
/*
* Transfer size is minimum of uio transfer
* and free space in pipe buffer.
*/
if (space > uio->uio_resid)
size = uio->uio_resid;
else
size = space;
/*
* First segment to transfer is minimum of
* transfer size and contiguous space in
* pipe buffer. If first segment to transfer
* is less than the transfer size, we've got
* a wraparound in the buffer.
*/
segsize = bp->size - bp->in;
if (segsize > size)
segsize = size;
/* Transfer first segment */
mutex_exit(lock);
error = uiomove((char *)bp->buffer + bp->in, segsize,
uio);
if (error == 0 && segsize < size) {
/*
* Transfer remaining part now, to
* support atomic writes. Wraparound
* happened.
*/
KASSERT(bp->in + segsize == bp->size);
error = uiomove(bp->buffer,
size - segsize, uio);
}
mutex_enter(lock);
if (error)
break;
bp->cnt += size;
KASSERT(bp->cnt <= bp->size);
wakeup_state = 0;
} else {
/*
* If the "read-side" has been blocked, wake it up now.
*/
cv_broadcast(&wpipe->pipe_rcv);
/*
* Don't block on non-blocking I/O.
*/
if (fp->f_flag & FNONBLOCK) {
error = EAGAIN;
break;
}
/*
* We have no more space and have something to offer,
* wake up select/poll.
*/
if (bp->cnt)
pipeselwakeup(wpipe, wpipe, POLL_IN);
if (wakeup_state & PIPE_RESTART) {
error = ERESTART;
break;
}
/*
* If read side wants to go away, we just issue a signal
* to ourselves.
*/
if (wpipe->pipe_state & PIPE_EOF) {
error = EPIPE;
break;
}
/*
* We implement a very minimal set of ioctls for compatibility with sockets.
*/
int
pipe_ioctl(file_t *fp, u_long cmd, void *data)
{
struct pipe *pipe = fp->f_pipe;
kmutex_t *lock = pipe->pipe_lock;
switch (cmd) {
case FIONBIO:
return (0);
case FIOASYNC:
mutex_enter(lock);
if (*(int *)data) {
pipe->pipe_state |= PIPE_ASYNC;
} else {
pipe->pipe_state &= ~PIPE_ASYNC;
}
mutex_exit(lock);
return (0);
case FIONREAD:
mutex_enter(lock);
*(int *)data = pipe->pipe_buffer.cnt;
mutex_exit(lock);
return (0);
case FIONWRITE:
/* Look at other side */
mutex_enter(lock);
pipe = pipe->pipe_peer;
if (pipe == NULL)
*(int *)data = 0;
else
*(int *)data = pipe->pipe_buffer.cnt;
mutex_exit(lock);
return (0);
case FIONSPACE:
/* Look at other side */
mutex_enter(lock);
pipe = pipe->pipe_peer;
if (pipe == NULL)
*(int *)data = 0;
else
*(int *)data = pipe->pipe_buffer.size -
pipe->pipe_buffer.cnt;
mutex_exit(lock);
return (0);
case TIOCSPGRP:
case FIOSETOWN:
return fsetown(&pipe->pipe_pgid, cmd, data);
case TIOCGPGRP:
case FIOGETOWN:
return fgetown(pipe->pipe_pgid, cmd, data);
}
return (EPASSTHROUGH);
}
int
pipe_poll(file_t *fp, int events)
{
struct pipe *rpipe = fp->f_pipe;
struct pipe *wpipe;
int eof = 0;
int revents = 0;
/*
* Unblock blocked reads/writes in order to allow close() to complete.
* System calls return ERESTART so that the fd is revalidated.
* (Partial writes return the transfer length.)
*/
mutex_enter(pipe->pipe_lock);
pipe->pipe_state |= PIPE_RESTART;
/* Wakeup both cvs, maybe we only need one, but maybe there are some
* other paths where wakeup is needed, and it saves deciding which! */
cv_broadcast(&pipe->pipe_rcv);
cv_broadcast(&pipe->pipe_wcv);
mutex_exit(pipe->pipe_lock);
}
static int
pipe_fpathconf(struct file *fp, int name, register_t *retval)
{
/*
* If the other side is blocked, wake it up saying that
* we want to close it down.
*/
pipe->pipe_state |= PIPE_EOF;
if (pipe->pipe_busy) {
while (pipe->pipe_busy) {
cv_broadcast(&pipe->pipe_wcv);
cv_wait_sig(&pipe->pipe_draincv, lock);
}
}
/*
* Any knote objects still left in the list are
* the one attached by peer. Since no one will
* traverse this list, we just clear it.
*
* XXX Exposes select/kqueue internals.
*/
SLIST_INIT(&pipe->pipe_sel.sel_klist);
