/*
* initialize and allocate VM and memory for pipe
*/
int
pipe_create(struct pipe *cpipe)
{
int error;
/* so pipe_free_kmem() doesn't follow junk pointer */
cpipe->pipe_buffer.buffer = NULL;
/*
* protect so pipeclose() doesn't follow a junk pointer
* if pipespace() fails.
*/
bzero(&cpipe->pipe_sel, sizeof cpipe->pipe_sel);
cpipe->pipe_state = 0;
cpipe->pipe_peer = NULL;
cpipe->pipe_busy = 0;
error = pipespace(cpipe, PIPE_SIZE);
if (error != 0)
return (error);
getnanotime(&cpipe->pipe_ctime);
cpipe->pipe_atime = cpipe->pipe_ctime;
cpipe->pipe_mtime = cpipe->pipe_ctime;
cpipe->pipe_pgid = NO_PID;
return (0);
}
/*
* expand the size of pipe while there is data to be read,
* and then free the old buffer once the current buffered
* data has been transferred to new storage.
* Required: PIPE_LOCK and io lock to be held by caller.
* returns 0 on success or no expansion possible
*/
static int
expand_pipespace(struct pipe *p, int target_size)
{
struct pipe tmp, oldpipe;
int error;
tmp.pipe_buffer.buffer = 0;
if (p->pipe_buffer.size >= (unsigned) target_size) {
return 0; /* the existing buffer is max size possible */
}
/* create enough space in the target */
error = pipespace(&tmp, target_size);
if (error != 0)
return (error);
oldpipe.pipe_buffer.buffer = p->pipe_buffer.buffer;
oldpipe.pipe_buffer.size = p->pipe_buffer.size;
memcpy(tmp.pipe_buffer.buffer, p->pipe_buffer.buffer, p->pipe_buffer.size);
if (p->pipe_buffer.cnt > 0 && p->pipe_buffer.in <= p->pipe_buffer.out ){
/* we are in State 3 and need extra copying for read to be consistent */
memcpy(&tmp.pipe_buffer.buffer[p->pipe_buffer.size], p->pipe_buffer.buffer, p->pipe_buffer.size);
p->pipe_buffer.in += p->pipe_buffer.size;
}
p->pipe_buffer.buffer = tmp.pipe_buffer.buffer;
p->pipe_buffer.size = tmp.pipe_buffer.size;
pipe_free_kmem(&oldpipe);
return 0;
}
int
pipe_write(struct file *fp, off_t *poff, struct uio *uio, struct ucred *cred)
{
int error = 0;
int orig_resid;
struct pipe *wpipe, *rpipe;
rpipe = (struct pipe *) fp->f_data;
wpipe = rpipe->pipe_peer;
/*
* detect loss of pipe read side, issue SIGPIPE if lost.
*/
if ((wpipe == NULL) || (wpipe->pipe_state & PIPE_EOF)) {
return (EPIPE);
}
++wpipe->pipe_busy;
/*
* If it is advantageous to resize the pipe buffer, do
* so.
*/
if ((uio->uio_resid > PIPE_SIZE) &&
(nbigpipe < LIMITBIGPIPES) &&
(wpipe->pipe_buffer.size <= PIPE_SIZE) &&
(wpipe->pipe_buffer.cnt == 0)) {
if ((error = pipelock(wpipe)) == 0) {
if (pipespace(wpipe, BIG_PIPE_SIZE) == 0)
nbigpipe++;
pipeunlock(wpipe);
}
}
/*
* If an early error occurred unbusy and return, waking up any pending
* readers.
*/
if (error) {
--wpipe->pipe_busy;
if ((wpipe->pipe_busy == 0) &&
(wpipe->pipe_state & PIPE_WANT)) {
wpipe->pipe_state &= ~(PIPE_WANT | PIPE_WANTR);
wakeup(wpipe);
}
return (error);
}
orig_resid = uio->uio_resid;
while (uio->uio_resid) {
int space;
retrywrite:
if (wpipe->pipe_state & PIPE_EOF) {
error = EPIPE;
break;
}
space = wpipe->pipe_buffer.size - wpipe->pipe_buffer.cnt;
/* Writes of size <= PIPE_BUF must be atomic. */
if ((space < uio->uio_resid) && (orig_resid <= PIPE_BUF))
space = 0;
if (space > 0) {
if ((error = pipelock(wpipe)) == 0) {
int size; /* Transfer size */
int segsize; /* first segment to transfer */
/*
* If a process blocked in uiomove, our
* value for space might be bad.
*
* XXX will we be ok if the reader has gone
* away here?
*/
if (space > wpipe->pipe_buffer.size -
wpipe->pipe_buffer.cnt) {
pipeunlock(wpipe);
goto retrywrite;
}
/*
* 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 = wpipe->pipe_buffer.size -
wpipe->pipe_buffer.in;
if (segsize > size)
segsize = size;
/* Transfer first segment */
error = uiomove(&wpipe->pipe_buffer.buffer[wpipe->pipe_buffer.in],
segsize, uio);
if (error == 0 && segsize < size) {
/*
* Transfer remaining part now, to
* support atomic writes. Wraparound
* happened.
*/
#ifdef DIAGNOSTIC
if (wpipe->pipe_buffer.in + segsize !=
wpipe->pipe_buffer.size)
panic("Expected pipe buffer wraparound disappeared");
#endif
error = uiomove(&wpipe->pipe_buffer.buffer[0],
size - segsize, uio);
}
if (error == 0) {
wpipe->pipe_buffer.in += size;
if (wpipe->pipe_buffer.in >=
wpipe->pipe_buffer.size) {
#ifdef DIAGNOSTIC
if (wpipe->pipe_buffer.in != size - segsize + wpipe->pipe_buffer.size)
panic("Expected wraparound bad");
#endif
wpipe->pipe_buffer.in = size - segsize;
}
wpipe->pipe_buffer.cnt += size;
#ifdef DIAGNOSTIC
if (wpipe->pipe_buffer.cnt > wpipe->pipe_buffer.size)
panic("Pipe buffer overflow");
#endif
}
pipeunlock(wpipe);
}
if (error)
break;
} else {
/*
* If the "read-side" has been blocked, wake it up now.
*/
if (wpipe->pipe_state & PIPE_WANTR) {
wpipe->pipe_state &= ~PIPE_WANTR;
wakeup(wpipe);
}
/*
* 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.
*/
pipeselwakeup(wpipe);
wpipe->pipe_state |= PIPE_WANTW;
error = tsleep(wpipe, (PRIBIO + 1)|PCATCH,
"pipewr", 0);
if (error)
break;
/*
* If read side wants to go away, we just issue a
* signal to ourselves.
*/
if (wpipe->pipe_state & PIPE_EOF) {
error = EPIPE;
break;
}
}
}
--wpipe->pipe_busy;
if ((wpipe->pipe_busy == 0) && (wpipe->pipe_state & PIPE_WANT)) {
wpipe->pipe_state &= ~(PIPE_WANT | PIPE_WANTR);
wakeup(wpipe);
} else if (wpipe->pipe_buffer.cnt > 0) {
/*
* If we have put any characters in the buffer, we wake up
* the reader.
*/
if (wpipe->pipe_state & PIPE_WANTR) {
wpipe->pipe_state &= ~PIPE_WANTR;
wakeup(wpipe);
}
}
/*
* Don't return EPIPE if I/O was successful
*/
if ((wpipe->pipe_buffer.cnt == 0) &&
(uio->uio_resid == 0) &&
(error == EPIPE)) {
error = 0;
}
if (error == 0)
getnanotime(&wpipe->pipe_mtime);
/*
* We have something to offer, wake up select/poll.
*/
if (wpipe->pipe_buffer.cnt)
pipeselwakeup(wpipe);
return (error);
}
/* ARGSUSED */
int
pipe(proc_t p, __unused struct pipe_args *uap, int32_t *retval)
{
struct fileproc *rf, *wf;
struct pipe *rpipe, *wpipe;
lck_mtx_t *pmtx;
int fd, error;
if ((pmtx = lck_mtx_alloc_init(pipe_mtx_grp, pipe_mtx_attr)) == NULL)
return (ENOMEM);
rpipe = wpipe = NULL;
if (pipe_create(&rpipe) || pipe_create(&wpipe)) {
error = ENFILE;
goto freepipes;
}
/*
* allocate the space for the normal I/O direction up
* front... we'll delay the allocation for the other
* direction until a write actually occurs (most likely it won't)...
*/
error = pipespace(rpipe, choose_pipespace(rpipe->pipe_buffer.size, 0));
if (error)
goto freepipes;
TAILQ_INIT(&rpipe->pipe_evlist);
TAILQ_INIT(&wpipe->pipe_evlist);
error = falloc(p, &rf, &fd, vfs_context_current());
if (error) {
goto freepipes;
}
retval[0] = fd;
/*
* for now we'll create half-duplex pipes(refer returns section above).
* this is what we've always supported..
*/
rf->f_flag = FREAD;
rf->f_data = (caddr_t)rpipe;
rf->f_ops = &pipeops;
error = falloc(p, &wf, &fd, vfs_context_current());
if (error) {
fp_free(p, retval[0], rf);
goto freepipes;
}
wf->f_flag = FWRITE;
wf->f_data = (caddr_t)wpipe;
wf->f_ops = &pipeops;
rpipe->pipe_peer = wpipe;
wpipe->pipe_peer = rpipe;
/* both structures share the same mutex */
rpipe->pipe_mtxp = wpipe->pipe_mtxp = pmtx;
retval[1] = fd;
#if CONFIG_MACF
/*
* XXXXXXXX SHOULD NOT HOLD FILE_LOCK() XXXXXXXXXXXX
*
* struct pipe represents a pipe endpoint. The MAC label is shared
* between the connected endpoints. As a result mac_pipe_label_init() and
* mac_pipe_label_associate() should only be called on one of the endpoints
* after they have been connected.
*/
mac_pipe_label_init(rpipe);
mac_pipe_label_associate(kauth_cred_get(), rpipe);
wpipe->pipe_label = rpipe->pipe_label;
#endif
proc_fdlock_spin(p);
procfdtbl_releasefd(p, retval[0], NULL);
procfdtbl_releasefd(p, retval[1], NULL);
fp_drop(p, retval[0], rf, 1);
fp_drop(p, retval[1], wf, 1);
proc_fdunlock(p);
return (0);
freepipes:
pipeclose(rpipe);
pipeclose(wpipe);
lck_mtx_free(pmtx, pipe_mtx_grp);
return (error);
}