The Design and Implementation of the FreeBSD Operating System, Second Edition
Now available: The Design and Implementation of the FreeBSD Operating System (Second Edition)


[ source navigation ] [ diff markup ] [ identifier search ] [ freetext search ] [ file search ] [ list types ] [ track identifier ]

FreeBSD/Linux Kernel Cross Reference
sys/kern/sys_pipe.c

Version: -  FREEBSD  -  FREEBSD-13-STABLE  -  FREEBSD-13-0  -  FREEBSD-12-STABLE  -  FREEBSD-12-0  -  FREEBSD-11-STABLE  -  FREEBSD-11-0  -  FREEBSD-10-STABLE  -  FREEBSD-10-0  -  FREEBSD-9-STABLE  -  FREEBSD-9-0  -  FREEBSD-8-STABLE  -  FREEBSD-8-0  -  FREEBSD-7-STABLE  -  FREEBSD-7-0  -  FREEBSD-6-STABLE  -  FREEBSD-6-0  -  FREEBSD-5-STABLE  -  FREEBSD-5-0  -  FREEBSD-4-STABLE  -  FREEBSD-3-STABLE  -  FREEBSD22  -  l41  -  OPENBSD  -  linux-2.6  -  MK84  -  PLAN9  -  xnu-8792 
SearchContext: -  none  -  3  -  10 

    1 /*      $NetBSD: sys_pipe.c,v 1.64.2.1 2005/09/14 20:35:05 tron Exp $   */
    2 
    3 /*-
    4  * Copyright (c) 2003 The NetBSD Foundation, Inc.
    5  * All rights reserved.
    6  *
    7  * This code is derived from software contributed to The NetBSD Foundation
    8  * by Paul Kranenburg.
    9  *
   10  * Redistribution and use in source and binary forms, with or without
   11  * modification, are permitted provided that the following conditions
   12  * are met:
   13  * 1. Redistributions of source code must retain the above copyright
   14  *    notice, this list of conditions and the following disclaimer.
   15  * 2. Redistributions in binary form must reproduce the above copyright
   16  *    notice, this list of conditions and the following disclaimer in the
   17  *    documentation and/or other materials provided with the distribution.
   18  * 3. All advertising materials mentioning features or use of this software
   19  *    must display the following acknowledgement:
   20  *        This product includes software developed by the NetBSD
   21  *        Foundation, Inc. and its contributors.
   22  * 4. Neither the name of The NetBSD Foundation nor the names of its
   23  *    contributors may be used to endorse or promote products derived
   24  *    from this software without specific prior written permission.
   25  *
   26  * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
   27  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
   28  * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
   29  * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
   30  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
   31  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
   32  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
   33  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
   34  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
   35  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
   36  * POSSIBILITY OF SUCH DAMAGE.
   37  */
   38 
   39 /*
   40  * Copyright (c) 1996 John S. Dyson
   41  * All rights reserved.
   42  *
   43  * Redistribution and use in source and binary forms, with or without
   44  * modification, are permitted provided that the following conditions
   45  * are met:
   46  * 1. Redistributions of source code must retain the above copyright
   47  *    notice immediately at the beginning of the file, without modification,
   48  *    this list of conditions, and the following disclaimer.
   49  * 2. Redistributions in binary form must reproduce the above copyright
   50  *    notice, this list of conditions and the following disclaimer in the
   51  *    documentation and/or other materials provided with the distribution.
   52  * 3. Absolutely no warranty of function or purpose is made by the author
   53  *    John S. Dyson.
   54  * 4. Modifications may be freely made to this file if the above conditions
   55  *    are met.
   56  *
   57  * $FreeBSD: src/sys/kern/sys_pipe.c,v 1.95 2002/03/09 22:06:31 alfred Exp $
   58  */
   59 
   60 /*
   61  * This file contains a high-performance replacement for the socket-based
   62  * pipes scheme originally used in FreeBSD/4.4Lite.  It does not support
   63  * all features of sockets, but does do everything that pipes normally
   64  * do.
   65  *
   66  * Adaption for NetBSD UVM, including uvm_loan() based direct write, was
   67  * written by Jaromir Dolecek.
   68  */
   69 
   70 /*
   71  * This code has two modes of operation, a small write mode and a large
   72  * write mode.  The small write mode acts like conventional pipes with
   73  * a kernel buffer.  If the buffer is less than PIPE_MINDIRECT, then the
   74  * "normal" pipe buffering is done.  If the buffer is between PIPE_MINDIRECT
   75  * and PIPE_SIZE in size it is mapped read-only into the kernel address space
   76  * using the UVM page loan facility from where the receiving process can copy
   77  * the data directly from the pages in the sending process.
   78  *
   79  * The constant PIPE_MINDIRECT is chosen to make sure that buffering will
   80  * happen for small transfers so that the system will not spend all of
   81  * its time context switching.  PIPE_SIZE is constrained by the
   82  * amount of kernel virtual memory.
   83  */
   84 
   85 #include <sys/cdefs.h>
   86 __KERNEL_RCSID(0, "$NetBSD: sys_pipe.c,v 1.64.2.1 2005/09/14 20:35:05 tron Exp $");
   87 
   88 #include <sys/param.h>
   89 #include <sys/systm.h>
   90 #include <sys/proc.h>
   91 #include <sys/fcntl.h>
   92 #include <sys/file.h>
   93 #include <sys/filedesc.h>
   94 #include <sys/filio.h>
   95 #include <sys/kernel.h>
   96 #include <sys/lock.h>
   97 #include <sys/ttycom.h>
   98 #include <sys/stat.h>
   99 #include <sys/malloc.h>
  100 #include <sys/poll.h>
  101 #include <sys/signalvar.h>
  102 #include <sys/vnode.h>
  103 #include <sys/uio.h>
  104 #include <sys/lock.h>
  105 #include <sys/select.h>
  106 #include <sys/mount.h>
  107 #include <sys/sa.h>
  108 #include <sys/syscallargs.h>
  109 #include <uvm/uvm.h>
  110 #include <sys/sysctl.h>
  111 #include <sys/kernel.h>
  112 
  113 #include <sys/pipe.h>
  114 
  115 /*
  116  * Avoid microtime(9), it's slow. We don't guard the read from time(9)
  117  * with splclock(9) since we don't actually need to be THAT sure the access
  118  * is atomic.
  119  */
  120 #define PIPE_TIMESTAMP(tvp)     (*(tvp) = time)
  121 
  122 
  123 /*
  124  * Use this define if you want to disable *fancy* VM things.  Expect an
  125  * approx 30% decrease in transfer rate.
  126  */
  127 /* #define PIPE_NODIRECT */
  128 
  129 /*
  130  * interfaces to the outside world
  131  */
  132 static int pipe_read(struct file *fp, off_t *offset, struct uio *uio,
  133                 struct ucred *cred, int flags);
  134 static int pipe_write(struct file *fp, off_t *offset, struct uio *uio,
  135                 struct ucred *cred, int flags);
  136 static int pipe_close(struct file *fp, struct proc *p);
  137 static int pipe_poll(struct file *fp, int events, struct proc *p);
  138 static int pipe_kqfilter(struct file *fp, struct knote *kn);
  139 static int pipe_stat(struct file *fp, struct stat *sb, struct proc *p);
  140 static int pipe_ioctl(struct file *fp, u_long cmd, void *data,
  141                 struct proc *p);
  142 
  143 static const struct fileops pipeops = {
  144         pipe_read, pipe_write, pipe_ioctl, fnullop_fcntl, pipe_poll,
  145         pipe_stat, pipe_close, pipe_kqfilter
  146 };
  147 
  148 /*
  149  * Default pipe buffer size(s), this can be kind-of large now because pipe
  150  * space is pageable.  The pipe code will try to maintain locality of
  151  * reference for performance reasons, so small amounts of outstanding I/O
  152  * will not wipe the cache.
  153  */
  154 #define MINPIPESIZE (PIPE_SIZE/3)
  155 #define MAXPIPESIZE (2*PIPE_SIZE/3)
  156 
  157 /*
  158  * Maximum amount of kva for pipes -- this is kind-of a soft limit, but
  159  * is there so that on large systems, we don't exhaust it.
  160  */
  161 #define MAXPIPEKVA (8*1024*1024)
  162 static int maxpipekva = MAXPIPEKVA;
  163 
  164 /*
  165  * Limit for direct transfers, we cannot, of course limit
  166  * the amount of kva for pipes in general though.
  167  */
  168 #define LIMITPIPEKVA (16*1024*1024)
  169 static int limitpipekva = LIMITPIPEKVA;
  170 
  171 /*
  172  * Limit the number of "big" pipes
  173  */
  174 #define LIMITBIGPIPES  32
  175 static int maxbigpipes = LIMITBIGPIPES;
  176 static int nbigpipe = 0;
  177 
  178 /*
  179  * Amount of KVA consumed by pipe buffers.
  180  */
  181 static int amountpipekva = 0;
  182 
  183 MALLOC_DEFINE(M_PIPE, "pipe", "Pipe structures");
  184 
  185 static void pipeclose(struct file *fp, struct pipe *pipe);
  186 static void pipe_free_kmem(struct pipe *pipe);
  187 static int pipe_create(struct pipe **pipep, int allockva);
  188 static int pipelock(struct pipe *pipe, int catch);
  189 static __inline void pipeunlock(struct pipe *pipe);
  190 static void pipeselwakeup(struct pipe *pipe, struct pipe *sigp, int code);
  191 #ifndef PIPE_NODIRECT
  192 static int pipe_direct_write(struct file *fp, struct pipe *wpipe,
  193     struct uio *uio);
  194 #endif
  195 static int pipespace(struct pipe *pipe, int size);
  196 
  197 #ifndef PIPE_NODIRECT
  198 static int pipe_loan_alloc(struct pipe *, int);
  199 static void pipe_loan_free(struct pipe *);
  200 #endif /* PIPE_NODIRECT */
  201 
  202 static POOL_INIT(pipe_pool, sizeof(struct pipe), 0, 0, 0, "pipepl",
  203     &pool_allocator_nointr);
  204 
  205 /*
  206  * The pipe system call for the DTYPE_PIPE type of pipes
  207  */
  208 
  209 /* ARGSUSED */
  210 int
  211 sys_pipe(l, v, retval)
  212         struct lwp *l;
  213         void *v;
  214         register_t *retval;
  215 {
  216         struct file *rf, *wf;
  217         struct pipe *rpipe, *wpipe;
  218         int fd, error;
  219         struct proc *p;
  220 
  221         p = l->l_proc;
  222         rpipe = wpipe = NULL;
  223         if (pipe_create(&rpipe, 1) || pipe_create(&wpipe, 0)) {
  224                 pipeclose(NULL, rpipe);
  225                 pipeclose(NULL, wpipe);
  226                 return (ENFILE);
  227         }
  228 
  229         /*
  230          * Note: the file structure returned from falloc() is marked
  231          * as 'larval' initially. Unless we mark it as 'mature' by
  232          * FILE_SET_MATURE(), any attempt to do anything with it would
  233          * return EBADF, including e.g. dup(2) or close(2). This avoids
  234          * file descriptor races if we block in the second falloc().
  235          */
  236 
  237         error = falloc(p, &rf, &fd);
  238         if (error)
  239                 goto free2;
  240         retval[0] = fd;
  241         rf->f_flag = FREAD;
  242         rf->f_type = DTYPE_PIPE;
  243         rf->f_data = (caddr_t)rpipe;
  244         rf->f_ops = &pipeops;
  245 
  246         error = falloc(p, &wf, &fd);
  247         if (error)
  248                 goto free3;
  249         retval[1] = fd;
  250         wf->f_flag = FWRITE;
  251         wf->f_type = DTYPE_PIPE;
  252         wf->f_data = (caddr_t)wpipe;
  253         wf->f_ops = &pipeops;
  254 
  255         rpipe->pipe_peer = wpipe;
  256         wpipe->pipe_peer = rpipe;
  257 
  258         FILE_SET_MATURE(rf);
  259         FILE_SET_MATURE(wf);
  260         FILE_UNUSE(rf, p);
  261         FILE_UNUSE(wf, p);
  262         return (0);
  263 free3:
  264         FILE_UNUSE(rf, p);
  265         ffree(rf);
  266         fdremove(p->p_fd, retval[0]);
  267 free2:
  268         pipeclose(NULL, wpipe);
  269         pipeclose(NULL, rpipe);
  270 
  271         return (error);
  272 }
  273 
  274 /*
  275  * Allocate kva for pipe circular buffer, the space is pageable
  276  * This routine will 'realloc' the size of a pipe safely, if it fails
  277  * it will retain the old buffer.
  278  * If it fails it will return ENOMEM.
  279  */
  280 static int
  281 pipespace(pipe, size)
  282         struct pipe *pipe;
  283         int size;
  284 {
  285         caddr_t buffer;
  286         /*
  287          * Allocate pageable virtual address space. Physical memory is
  288          * allocated on demand.
  289          */
  290         buffer = (caddr_t) uvm_km_valloc(kernel_map, round_page(size));
  291         if (buffer == NULL)
  292                 return (ENOMEM);
  293 
  294         /* free old resources if we're resizing */
  295         pipe_free_kmem(pipe);
  296         pipe->pipe_buffer.buffer = buffer;
  297         pipe->pipe_buffer.size = size;
  298         pipe->pipe_buffer.in = 0;
  299         pipe->pipe_buffer.out = 0;
  300         pipe->pipe_buffer.cnt = 0;
  301         amountpipekva += pipe->pipe_buffer.size;
  302         return (0);
  303 }
  304 
  305 /*
  306  * Initialize and allocate VM and memory for pipe.
  307  */
  308 static int
  309 pipe_create(pipep, allockva)
  310         struct pipe **pipep;
  311         int allockva;
  312 {
  313         struct pipe *pipe;
  314         int error;
  315 
  316         pipe = *pipep = pool_get(&pipe_pool, PR_WAITOK);
  317 
  318         /* Initialize */
  319         memset(pipe, 0, sizeof(struct pipe));
  320         pipe->pipe_state = PIPE_SIGNALR;
  321 
  322         PIPE_TIMESTAMP(&pipe->pipe_ctime);
  323         pipe->pipe_atime = pipe->pipe_ctime;
  324         pipe->pipe_mtime = pipe->pipe_ctime;
  325         simple_lock_init(&pipe->pipe_slock);
  326         lockinit(&pipe->pipe_lock, PSOCK | PCATCH, "pipelk", 0, 0);
  327 
  328         if (allockva && (error = pipespace(pipe, PIPE_SIZE)))
  329                 return (error);
  330 
  331         return (0);
  332 }
  333 
  334 
  335 /*
  336  * Lock a pipe for I/O, blocking other access
  337  * Called with pipe spin lock held.
  338  * Return with pipe spin lock released on success.
  339  */
  340 static int
  341 pipelock(pipe, catch)
  342         struct pipe *pipe;
  343         int catch;
  344 {
  345         int error;
  346 
  347         LOCK_ASSERT(simple_lock_held(&pipe->pipe_slock));
  348 
  349         while (1) {
  350                 error = lockmgr(&pipe->pipe_lock, LK_EXCLUSIVE | LK_INTERLOCK,
  351                                 &pipe->pipe_slock);
  352                 if (error == 0)
  353                         break;
  354 
  355                 simple_lock(&pipe->pipe_slock);
  356                 if (catch || (error != EINTR && error != ERESTART))
  357                         break;
  358                 /*
  359                  * XXX XXX XXX
  360                  * The pipe lock is initialised with PCATCH on and we cannot
  361                  * override this in a lockmgr() call. Thus a pending signal
  362                  * will cause lockmgr() to return with EINTR or ERESTART.
  363                  * We cannot simply re-enter lockmgr() at this point since
  364                  * the pending signals have not yet been posted and would
  365                  * cause an immediate EINTR/ERESTART return again.
  366                  * As a workaround we pause for a while here, giving the lock
  367                  * a chance to drain, before trying again.
  368                  * XXX XXX XXX
  369                  *
  370                  * NOTE: Consider dropping PCATCH from this lock; in practice
  371                  * it is never held for long enough periods for having it
  372                  * interruptable at the start of pipe_read/pipe_write to be
  373                  * beneficial.
  374                  */
  375                 (void) ltsleep(&lbolt, PSOCK, "rstrtpipelock", hz,
  376                     &pipe->pipe_slock);
  377         }
  378         return (error);
  379 }
  380 
  381 /*
  382  * unlock a pipe I/O lock
  383  */
  384 static __inline void
  385 pipeunlock(pipe)
  386         struct pipe *pipe;
  387 {
  388 
  389         lockmgr(&pipe->pipe_lock, LK_RELEASE, NULL);
  390 }
  391 
  392 /*
  393  * Select/poll wakup. This also sends SIGIO to peer connected to
  394  * 'sigpipe' side of pipe.
  395  */
  396 static void
  397 pipeselwakeup(selp, sigp, code)
  398         struct pipe *selp, *sigp;
  399         int code;
  400 {
  401         int band;
  402 
  403         selnotify(&selp->pipe_sel, NOTE_SUBMIT);
  404 
  405         if (sigp == NULL || (sigp->pipe_state & PIPE_ASYNC) == 0)
  406                 return;
  407 
  408         switch (code) {
  409         case POLL_IN:
  410                 band = POLLIN|POLLRDNORM;
  411                 break;
  412         case POLL_OUT:
  413                 band = POLLOUT|POLLWRNORM;
  414                 break;
  415         case POLL_HUP:
  416                 band = POLLHUP;
  417                 break;
  418 #if POLL_HUP != POLL_ERR
  419         case POLL_ERR:
  420                 band = POLLERR;
  421                 break;
  422 #endif
  423         default:
  424                 band = 0;
  425 #ifdef DIAGNOSTIC
  426                 printf("bad siginfo code %d in pipe notification.\n", code);
  427 #endif
  428                 break;
  429         }
  430 
  431         fownsignal(sigp->pipe_pgid, SIGIO, code, band, selp);
  432 }
  433 
  434 /* ARGSUSED */
  435 static int
  436 pipe_read(fp, offset, uio, cred, flags)
  437         struct file *fp;
  438         off_t *offset;
  439         struct uio *uio;
  440         struct ucred *cred;
  441         int flags;
  442 {
  443         struct pipe *rpipe = (struct pipe *) fp->f_data;
  444         struct pipebuf *bp = &rpipe->pipe_buffer;
  445         int error;
  446         size_t nread = 0;
  447         size_t size;
  448         size_t ocnt;
  449 
  450         PIPE_LOCK(rpipe);
  451         ++rpipe->pipe_busy;
  452         ocnt = bp->cnt;
  453 
  454 again:
  455         error = pipelock(rpipe, 1);
  456         if (error)
  457                 goto unlocked_error;
  458 
  459         while (uio->uio_resid) {
  460                 /*
  461                  * normal pipe buffer receive
  462                  */
  463                 if (bp->cnt > 0) {
  464                         size = bp->size - bp->out;
  465                         if (size > bp->cnt)
  466                                 size = bp->cnt;
  467                         if (size > uio->uio_resid)
  468                                 size = uio->uio_resid;
  469 
  470                         error = uiomove(&bp->buffer[bp->out], size, uio);
  471                         if (error)
  472                                 break;
  473 
  474                         bp->out += size;
  475                         if (bp->out >= bp->size)
  476                                 bp->out = 0;
  477 
  478                         bp->cnt -= size;
  479 
  480                         /*
  481                          * If there is no more to read in the pipe, reset
  482                          * its pointers to the beginning.  This improves
  483                          * cache hit stats.
  484                          */
  485                         if (bp->cnt == 0) {
  486                                 bp->in = 0;
  487                                 bp->out = 0;
  488                         }
  489                         nread += size;
  490 #ifndef PIPE_NODIRECT
  491                 } else if ((rpipe->pipe_state & PIPE_DIRECTR) != 0) {
  492                         /*
  493                          * Direct copy, bypassing a kernel buffer.
  494                          */
  495                         caddr_t va;
  496 
  497                         KASSERT(rpipe->pipe_state & PIPE_DIRECTW);
  498 
  499                         size = rpipe->pipe_map.cnt;
  500                         if (size > uio->uio_resid)
  501                                 size = uio->uio_resid;
  502 
  503                         va = (caddr_t) rpipe->pipe_map.kva +
  504                             rpipe->pipe_map.pos;
  505                         error = uiomove(va, size, uio);
  506                         if (error)
  507                                 break;
  508                         nread += size;
  509                         rpipe->pipe_map.pos += size;
  510                         rpipe->pipe_map.cnt -= size;
  511                         if (rpipe->pipe_map.cnt == 0) {
  512                                 PIPE_LOCK(rpipe);
  513                                 rpipe->pipe_state &= ~PIPE_DIRECTR;
  514                                 wakeup(rpipe);
  515                                 PIPE_UNLOCK(rpipe);
  516                         }
  517 #endif
  518                 } else {
  519                         /*
  520                          * Break if some data was read.
  521                          */
  522                         if (nread > 0)
  523                                 break;
  524 
  525                         PIPE_LOCK(rpipe);
  526 
  527                         /*
  528                          * detect EOF condition
  529                          * read returns 0 on EOF, no need to set error
  530                          */
  531                         if (rpipe->pipe_state & PIPE_EOF) {
  532                                 PIPE_UNLOCK(rpipe);
  533                                 break;
  534                         }
  535 
  536                         /*
  537                          * don't block on non-blocking I/O
  538                          */
  539                         if (fp->f_flag & FNONBLOCK) {
  540                                 PIPE_UNLOCK(rpipe);
  541                                 error = EAGAIN;
  542                                 break;
  543                         }
  544 
  545                         /*
  546                          * Unlock the pipe buffer for our remaining processing.
  547                          * We will either break out with an error or we will
  548                          * sleep and relock to loop.
  549                          */
  550                         pipeunlock(rpipe);
  551 
  552                         /*
  553                          * The PIPE_DIRECTR flag is not under the control
  554                          * of the long-term lock (see pipe_direct_write()),
  555                          * so re-check now while holding the spin lock.
  556                          */
  557                         if ((rpipe->pipe_state & PIPE_DIRECTR) != 0)
  558                                 goto again;
  559 
  560                         /*
  561                          * We want to read more, wake up select/poll.
  562                          */
  563                         pipeselwakeup(rpipe, rpipe->pipe_peer, POLL_IN);
  564 
  565                         /*
  566                          * If the "write-side" is blocked, wake it up now.
  567                          */
  568                         if (rpipe->pipe_state & PIPE_WANTW) {
  569                                 rpipe->pipe_state &= ~PIPE_WANTW;
  570                                 wakeup(rpipe);
  571                         }
  572 
  573                         /* Now wait until the pipe is filled */
  574                         rpipe->pipe_state |= PIPE_WANTR;
  575                         error = ltsleep(rpipe, PSOCK | PCATCH,
  576                                         "piperd", 0, &rpipe->pipe_slock);
  577                         if (error != 0)
  578                                 goto unlocked_error;
  579                         goto again;
  580                 }
  581         }
  582 
  583         if (error == 0)
  584                 PIPE_TIMESTAMP(&rpipe->pipe_atime);
  585 
  586         PIPE_LOCK(rpipe);
  587         pipeunlock(rpipe);
  588 
  589 unlocked_error:
  590         --rpipe->pipe_busy;
  591 
  592         /*
  593          * PIPE_WANTCLOSE processing only makes sense if pipe_busy is 0.
  594          */
  595         if ((rpipe->pipe_busy == 0) && (rpipe->pipe_state & PIPE_WANTCLOSE)) {
  596                 rpipe->pipe_state &= ~(PIPE_WANTCLOSE|PIPE_WANTW);
  597                 wakeup(rpipe);
  598         } else if (bp->cnt < MINPIPESIZE) {
  599                 /*
  600                  * Handle write blocking hysteresis.
  601                  */
  602                 if (rpipe->pipe_state & PIPE_WANTW) {
  603                         rpipe->pipe_state &= ~PIPE_WANTW;
  604                         wakeup(rpipe);
  605                 }
  606         }
  607 
  608         /*
  609          * If anything was read off the buffer, signal to the writer it's
  610          * possible to write more data. Also send signal if we are here for the
  611          * first time after last write.
  612          */
  613         if ((bp->size - bp->cnt) >= PIPE_BUF
  614             && (ocnt != bp->cnt || (rpipe->pipe_state & PIPE_SIGNALR))) {
  615                 pipeselwakeup(rpipe, rpipe->pipe_peer, POLL_OUT);
  616                 rpipe->pipe_state &= ~PIPE_SIGNALR;
  617         }
  618 
  619         PIPE_UNLOCK(rpipe);
  620         return (error);
  621 }
  622 
  623 #ifndef PIPE_NODIRECT
  624 /*
  625  * Allocate structure for loan transfer.
  626  */
  627 static int
  628 pipe_loan_alloc(wpipe, npages)
  629         struct pipe *wpipe;
  630         int npages;
  631 {
  632         vsize_t len;
  633 
  634         len = (vsize_t)npages << PAGE_SHIFT;
  635         wpipe->pipe_map.kva = uvm_km_valloc_wait(kernel_map, len);
  636         if (wpipe->pipe_map.kva == 0)
  637                 return (ENOMEM);
  638 
  639         amountpipekva += len;
  640         wpipe->pipe_map.npages = npages;
  641         wpipe->pipe_map.pgs = malloc(npages * sizeof(struct vm_page *), M_PIPE,
  642             M_WAITOK);
  643         return (0);
  644 }
  645 
  646 /*
  647  * Free resources allocated for loan transfer.
  648  */
  649 static void
  650 pipe_loan_free(wpipe)
  651         struct pipe *wpipe;
  652 {
  653         vsize_t len;
  654 
  655         len = (vsize_t)wpipe->pipe_map.npages << PAGE_SHIFT;
  656         uvm_km_free(kernel_map, wpipe->pipe_map.kva, len);
  657         wpipe->pipe_map.kva = 0;
  658         amountpipekva -= len;
  659         free(wpipe->pipe_map.pgs, M_PIPE);
  660         wpipe->pipe_map.pgs = NULL;
  661 }
  662 
  663 /*
  664  * NetBSD direct write, using uvm_loan() mechanism.
  665  * This implements the pipe buffer write mechanism.  Note that only
  666  * a direct write OR a normal pipe write can be pending at any given time.
  667  * If there are any characters in the pipe buffer, the direct write will
  668  * be deferred until the receiving process grabs all of the bytes from
  669  * the pipe buffer.  Then the direct mapping write is set-up.
  670  *
  671  * Called with the long-term pipe lock held.
  672  */
  673 static int
  674 pipe_direct_write(fp, wpipe, uio)
  675         struct file *fp;
  676         struct pipe *wpipe;
  677         struct uio *uio;
  678 {
  679         int error, npages, j;
  680         struct vm_page **pgs;
  681         vaddr_t bbase, kva, base, bend;
  682         vsize_t blen, bcnt;
  683         voff_t bpos;
  684 
  685         KASSERT(wpipe->pipe_map.cnt == 0);
  686 
  687         /*
  688          * Handle first PIPE_CHUNK_SIZE bytes of buffer. Deal with buffers
  689          * not aligned to PAGE_SIZE.
  690          */
  691         bbase = (vaddr_t)uio->uio_iov->iov_base;
  692         base = trunc_page(bbase);
  693         bend = round_page(bbase + uio->uio_iov->iov_len);
  694         blen = bend - base;
  695         bpos = bbase - base;
  696 
  697         if (blen > PIPE_DIRECT_CHUNK) {
  698                 blen = PIPE_DIRECT_CHUNK;
  699                 bend = base + blen;
  700                 bcnt = PIPE_DIRECT_CHUNK - bpos;
  701         } else {
  702                 bcnt = uio->uio_iov->iov_len;
  703         }
  704         npages = blen >> PAGE_SHIFT;
  705 
  706         /*
  707          * Free the old kva if we need more pages than we have
  708          * allocated.
  709          */
  710         if (wpipe->pipe_map.kva != 0 && npages > wpipe->pipe_map.npages)
  711                 pipe_loan_free(wpipe);
  712 
  713         /* Allocate new kva. */
  714         if (wpipe->pipe_map.kva == 0) {
  715                 error = pipe_loan_alloc(wpipe, npages);
  716                 if (error)
  717                         return (error);
  718         }
  719 
  720         /* Loan the write buffer memory from writer process */
  721         pgs = wpipe->pipe_map.pgs;
  722         error = uvm_loan(&uio->uio_procp->p_vmspace->vm_map, base, blen,
  723                          pgs, UVM_LOAN_TOPAGE);
  724         if (error) {
  725                 pipe_loan_free(wpipe);
  726                 return (ENOMEM); /* so that caller fallback to ordinary write */
  727         }
  728 
  729         /* Enter the loaned pages to kva */
  730         kva = wpipe->pipe_map.kva;
  731         for (j = 0; j < npages; j++, kva += PAGE_SIZE) {
  732                 pmap_kenter_pa(kva, VM_PAGE_TO_PHYS(pgs[j]), VM_PROT_READ);
  733         }
  734         pmap_update(pmap_kernel());
  735 
  736         /* Now we can put the pipe in direct write mode */
  737         wpipe->pipe_map.pos = bpos;
  738         wpipe->pipe_map.cnt = bcnt;
  739         wpipe->pipe_state |= PIPE_DIRECTW;
  740 
  741         /*
  742          * But before we can let someone do a direct read,
  743          * we have to wait until the pipe is drained.
  744          */
  745 
  746         /* Relase the pipe lock while we wait */
  747         PIPE_LOCK(wpipe);
  748         pipeunlock(wpipe);
  749 
  750         while (error == 0 && wpipe->pipe_buffer.cnt > 0) {
  751                 if (wpipe->pipe_state & PIPE_WANTR) {
  752                         wpipe->pipe_state &= ~PIPE_WANTR;
  753                         wakeup(wpipe);
  754                 }
  755 
  756                 wpipe->pipe_state |= PIPE_WANTW;
  757                 error = ltsleep(wpipe, PSOCK | PCATCH, "pipdwc", 0,
  758                                 &wpipe->pipe_slock);
  759                 if (error == 0 && wpipe->pipe_state & PIPE_EOF)
  760                         error = EPIPE;
  761         }
  762 
  763         /* Pipe is drained; next read will off the direct buffer */
  764         wpipe->pipe_state |= PIPE_DIRECTR;
  765 
  766         /* Wait until the reader is done */
  767         while (error == 0 && (wpipe->pipe_state & PIPE_DIRECTR)) {
  768                 if (wpipe->pipe_state & PIPE_WANTR) {
  769                         wpipe->pipe_state &= ~PIPE_WANTR;
  770                         wakeup(wpipe);
  771                 }
  772                 pipeselwakeup(wpipe, wpipe, POLL_IN);
  773                 error = ltsleep(wpipe, PSOCK | PCATCH, "pipdwt", 0,
  774                                 &wpipe->pipe_slock);
  775                 if (error == 0 && wpipe->pipe_state & PIPE_EOF)
  776                         error = EPIPE;
  777         }
  778 
  779         /* Take pipe out of direct write mode */
  780         wpipe->pipe_state &= ~(PIPE_DIRECTW | PIPE_DIRECTR);
  781 
  782         /* Acquire the pipe lock and cleanup */
  783         (void)pipelock(wpipe, 0);
  784         if (pgs != NULL) {
  785                 pmap_kremove(wpipe->pipe_map.kva, blen);
  786                 uvm_unloan(pgs, npages, UVM_LOAN_TOPAGE);
  787         }
  788         if (error || amountpipekva > maxpipekva)
  789                 pipe_loan_free(wpipe);
  790 
  791         if (error) {
  792                 pipeselwakeup(wpipe, wpipe, POLL_ERR);
  793 
  794                 /*
  795                  * If nothing was read from what we offered, return error
  796                  * straight on. Otherwise update uio resid first. Caller
  797                  * will deal with the error condition, returning short
  798                  * write, error, or restarting the write(2) as appropriate.
  799                  */
  800                 if (wpipe->pipe_map.cnt == bcnt) {
  801                         wpipe->pipe_map.cnt = 0;
  802                         wakeup(wpipe);
  803                         return (error);
  804                 }
  805 
  806                 bcnt -= wpipe->pipe_map.cnt;
  807         }
  808 
  809         uio->uio_resid -= bcnt;
  810         /* uio_offset not updated, not set/used for write(2) */
  811         uio->uio_iov->iov_base = (char *)uio->uio_iov->iov_base + bcnt;
  812         uio->uio_iov->iov_len -= bcnt;
  813         if (uio->uio_iov->iov_len == 0) {
  814                 uio->uio_iov++;
  815                 uio->uio_iovcnt--;
  816         }
  817 
  818         wpipe->pipe_map.cnt = 0;
  819         return (error);
  820 }
  821 #endif /* !PIPE_NODIRECT */
  822 
  823 static int
  824 pipe_write(fp, offset, uio, cred, flags)
  825         struct file *fp;
  826         off_t *offset;
  827         struct uio *uio;
  828         struct ucred *cred;
  829         int flags;
  830 {
  831         struct pipe *wpipe, *rpipe;
  832         struct pipebuf *bp;
  833         int error;
  834 
  835         /* We want to write to our peer */
  836         rpipe = (struct pipe *) fp->f_data;
  837 
  838 retry:
  839         error = 0;
  840         PIPE_LOCK(rpipe);
  841         wpipe = rpipe->pipe_peer;
  842 
  843         /*
  844          * Detect loss of pipe read side, issue SIGPIPE if lost.
  845          */
  846         if (wpipe == NULL)
  847                 error = EPIPE;
  848         else if (simple_lock_try(&wpipe->pipe_slock) == 0) {
  849                 /* Deal with race for peer */
  850                 PIPE_UNLOCK(rpipe);
  851                 goto retry;
  852         } else if ((wpipe->pipe_state & PIPE_EOF) != 0) {
  853                 PIPE_UNLOCK(wpipe);
  854                 error = EPIPE;
  855         }
  856 
  857         PIPE_UNLOCK(rpipe);
  858         if (error != 0)
  859                 return (error);
  860 
  861         ++wpipe->pipe_busy;
  862 
  863         /* Aquire the long-term pipe lock */
  864         if ((error = pipelock(wpipe,1)) != 0) {
  865                 --wpipe->pipe_busy;
  866                 if (wpipe->pipe_busy == 0
  867                     && (wpipe->pipe_state & PIPE_WANTCLOSE)) {
  868                         wpipe->pipe_state &= ~(PIPE_WANTCLOSE | PIPE_WANTR);
  869                         wakeup(wpipe);
  870                 }
  871                 PIPE_UNLOCK(wpipe);
  872                 return (error);
  873         }
  874 
  875         bp = &wpipe->pipe_buffer;
  876 
  877         /*
  878          * If it is advantageous to resize the pipe buffer, do so.
  879          */
  880         if ((uio->uio_resid > PIPE_SIZE) &&
  881             (nbigpipe < maxbigpipes) &&
  882 #ifndef PIPE_NODIRECT
  883             (wpipe->pipe_state & PIPE_DIRECTW) == 0 &&
  884 #endif
  885             (bp->size <= PIPE_SIZE) && (bp->cnt == 0)) {
  886 
  887                 if (pipespace(wpipe, BIG_PIPE_SIZE) == 0)
  888                         nbigpipe++;
  889         }
  890 
  891         while (uio->uio_resid) {
  892                 size_t space;
  893 
  894 #ifndef PIPE_NODIRECT
  895                 /*
  896                  * Pipe buffered writes cannot be coincidental with
  897                  * direct writes.  Also, only one direct write can be
  898                  * in progress at any one time.  We wait until the currently
  899                  * executing direct write is completed before continuing.
  900                  *
  901                  * We break out if a signal occurs or the reader goes away.
  902                  */
  903                 while (error == 0 && wpipe->pipe_state & PIPE_DIRECTW) {
  904                         PIPE_LOCK(wpipe);
  905                         if (wpipe->pipe_state & PIPE_WANTR) {
  906                                 wpipe->pipe_state &= ~PIPE_WANTR;
  907                                 wakeup(wpipe);
  908                         }
  909                         pipeunlock(wpipe);
  910                         error = ltsleep(wpipe, PSOCK | PCATCH,
  911                                         "pipbww", 0, &wpipe->pipe_slock);
  912 
  913                         (void)pipelock(wpipe, 0);
  914                         if (wpipe->pipe_state & PIPE_EOF)
  915                                 error = EPIPE;
  916                 }
  917                 if (error)
  918                         break;
  919 
  920                 /*
  921                  * If the transfer is large, we can gain performance if
  922                  * we do process-to-process copies directly.
  923                  * If the write is non-blocking, we don't use the
  924                  * direct write mechanism.
  925                  *
  926                  * The direct write mechanism will detect the reader going
  927                  * away on us.
  928                  */
  929                 if ((uio->uio_iov->iov_len >= PIPE_MINDIRECT) &&
  930                     (fp->f_flag & FNONBLOCK) == 0 &&
  931                     (wpipe->pipe_map.kva || (amountpipekva < limitpipekva))) {
  932                         error = pipe_direct_write(fp, wpipe, uio);
  933 
  934                         /*
  935                          * Break out if error occurred, unless it's ENOMEM.
  936                          * ENOMEM means we failed to allocate some resources
  937                          * for direct write, so we just fallback to ordinary
  938                          * write. If the direct write was successful,
  939                          * process rest of data via ordinary write.
  940                          */
  941                         if (error == 0)
  942                                 continue;
  943 
  944                         if (error != ENOMEM)
  945                                 break;
  946                 }
  947 #endif /* PIPE_NODIRECT */
  948 
  949                 space = bp->size - bp->cnt;
  950 
  951                 /* Writes of size <= PIPE_BUF must be atomic. */
  952                 if ((space < uio->uio_resid) && (uio->uio_resid <= PIPE_BUF))
  953                         space = 0;
  954 
  955                 if (space > 0) {
  956                         int size;       /* Transfer size */
  957                         int segsize;    /* first segment to transfer */
  958 
  959                         /*
  960                          * Transfer size is minimum of uio transfer
  961                          * and free space in pipe buffer.
  962                          */
  963                         if (space > uio->uio_resid)
  964                                 size = uio->uio_resid;
  965                         else
  966                                 size = space;
  967                         /*
  968                          * First segment to transfer is minimum of
  969                          * transfer size and contiguous space in
  970                          * pipe buffer.  If first segment to transfer
  971                          * is less than the transfer size, we've got
  972                          * a wraparound in the buffer.
  973                          */
  974                         segsize = bp->size - bp->in;
  975                         if (segsize > size)
  976                                 segsize = size;
  977 
  978                         /* Transfer first segment */
  979                         error = uiomove(&bp->buffer[bp->in], segsize, uio);
  980 
  981                         if (error == 0 && segsize < size) {
  982                                 /*
  983                                  * Transfer remaining part now, to
  984                                  * support atomic writes.  Wraparound
  985                                  * happened.
  986                                  */
  987 #ifdef DEBUG
  988                                 if (bp->in + segsize != bp->size)
  989                                         panic("Expected pipe buffer wraparound disappeared");
  990 #endif
  991 
  992                                 error = uiomove(&bp->buffer[0],
  993                                                 size - segsize, uio);
  994                         }
  995                         if (error)
  996                                 break;
  997 
  998                         bp->in += size;
  999                         if (bp->in >= bp->size) {
 1000 #ifdef DEBUG
 1001                                 if (bp->in != size - segsize + bp->size)
 1002                                         panic("Expected wraparound bad");
 1003 #endif
 1004                                 bp->in = size - segsize;
 1005                         }
 1006 
 1007                         bp->cnt += size;
 1008 #ifdef DEBUG
 1009                         if (bp->cnt > bp->size)
 1010                                 panic("Pipe buffer overflow");
 1011 #endif
 1012                 } else {
 1013                         /*
 1014                          * If the "read-side" has been blocked, wake it up now.
 1015                          */
 1016                         PIPE_LOCK(wpipe);
 1017                         if (wpipe->pipe_state & PIPE_WANTR) {
 1018                                 wpipe->pipe_state &= ~PIPE_WANTR;
 1019                                 wakeup(wpipe);
 1020                         }
 1021                         PIPE_UNLOCK(wpipe);
 1022 
 1023                         /*
 1024                          * don't block on non-blocking I/O
 1025                          */
 1026                         if (fp->f_flag & FNONBLOCK) {
 1027                                 error = EAGAIN;
 1028                                 break;
 1029                         }
 1030 
 1031                         /*
 1032                          * We have no more space and have something to offer,
 1033                          * wake up select/poll.
 1034                          */
 1035                         if (bp->cnt)
 1036                                 pipeselwakeup(wpipe, wpipe, POLL_OUT);
 1037 
 1038                         PIPE_LOCK(wpipe);
 1039                         pipeunlock(wpipe);
 1040                         wpipe->pipe_state |= PIPE_WANTW;
 1041                         error = ltsleep(wpipe, PSOCK | PCATCH, "pipewr", 0,
 1042                                         &wpipe->pipe_slock);
 1043                         (void)pipelock(wpipe, 0);
 1044                         if (error != 0)
 1045                                 break;
 1046                         /*
 1047                          * If read side wants to go away, we just issue a signal
 1048                          * to ourselves.
 1049                          */
 1050                         if (wpipe->pipe_state & PIPE_EOF) {
 1051                                 error = EPIPE;
 1052                                 break;
 1053                         }
 1054                 }
 1055         }
 1056 
 1057         PIPE_LOCK(wpipe);
 1058         --wpipe->pipe_busy;
 1059         if ((wpipe->pipe_busy == 0) && (wpipe->pipe_state & PIPE_WANTCLOSE)) {
 1060                 wpipe->pipe_state &= ~(PIPE_WANTCLOSE | PIPE_WANTR);
 1061                 wakeup(wpipe);
 1062         } else if (bp->cnt > 0) {
 1063                 /*
 1064                  * If we have put any characters in the buffer, we wake up
 1065                  * the reader.
 1066                  */
 1067                 if (wpipe->pipe_state & PIPE_WANTR) {
 1068                         wpipe->pipe_state &= ~PIPE_WANTR;
 1069                         wakeup(wpipe);
 1070                 }
 1071         }
 1072 
 1073         /*
 1074          * Don't return EPIPE if I/O was successful
 1075          */
 1076         if (error == EPIPE && bp->cnt == 0 && uio->uio_resid == 0)
 1077                 error = 0;
 1078 
 1079         if (error == 0)
 1080                 PIPE_TIMESTAMP(&wpipe->pipe_mtime);
 1081 
 1082         /*
 1083          * We have something to offer, wake up select/poll.
 1084          * wpipe->pipe_map.cnt is always 0 in this point (direct write
 1085          * is only done synchronously), so check only wpipe->pipe_buffer.cnt
 1086          */
 1087         if (bp->cnt)
 1088                 pipeselwakeup(wpipe, wpipe, POLL_OUT);
 1089 
 1090         /*
 1091          * Arrange for next read(2) to do a signal.
 1092          */
 1093         wpipe->pipe_state |= PIPE_SIGNALR;
 1094 
 1095         pipeunlock(wpipe);
 1096         PIPE_UNLOCK(wpipe);
 1097         return (error);
 1098 }
 1099 
 1100 /*
 1101  * we implement a very minimal set of ioctls for compatibility with sockets.
 1102  */
 1103 int
 1104 pipe_ioctl(fp, cmd, data, p)
 1105         struct file *fp;
 1106         u_long cmd;
 1107         void *data;
 1108         struct proc *p;
 1109 {
 1110         struct pipe *pipe = (struct pipe *)fp->f_data;
 1111 
 1112         switch (cmd) {
 1113 
 1114         case FIONBIO:
 1115                 return (0);
 1116 
 1117         case FIOASYNC:
 1118                 PIPE_LOCK(pipe);
 1119                 if (*(int *)data) {
 1120                         pipe->pipe_state |= PIPE_ASYNC;
 1121                 } else {
 1122                         pipe->pipe_state &= ~PIPE_ASYNC;
 1123                 }
 1124                 PIPE_UNLOCK(pipe);
 1125                 return (0);
 1126 
 1127         case FIONREAD:
 1128                 PIPE_LOCK(pipe);
 1129 #ifndef PIPE_NODIRECT
 1130                 if (pipe->pipe_state & PIPE_DIRECTW)
 1131                         *(int *)data = pipe->pipe_map.cnt;
 1132                 else
 1133 #endif
 1134                         *(int *)data = pipe->pipe_buffer.cnt;
 1135                 PIPE_UNLOCK(pipe);
 1136                 return (0);
 1137 
 1138         case FIONWRITE:
 1139                 /* Look at other side */
 1140                 pipe = pipe->pipe_peer;
 1141                 PIPE_LOCK(pipe);
 1142 #ifndef PIPE_NODIRECT
 1143                 if (pipe->pipe_state & PIPE_DIRECTW)
 1144                         *(int *)data = pipe->pipe_map.cnt;
 1145                 else
 1146 #endif
 1147                         *(int *)data = pipe->pipe_buffer.cnt;
 1148                 PIPE_UNLOCK(pipe);
 1149                 return (0);
 1150 
 1151         case FIONSPACE:
 1152                 /* Look at other side */
 1153                 pipe = pipe->pipe_peer;
 1154                 PIPE_LOCK(pipe);
 1155 #ifndef PIPE_NODIRECT
 1156                 /*
 1157                  * If we're in direct-mode, we don't really have a
 1158                  * send queue, and any other write will block. Thus
 1159                  * zero seems like the best answer.
 1160                  */
 1161                 if (pipe->pipe_state & PIPE_DIRECTW)
 1162                         *(int *)data = 0;
 1163                 else
 1164 #endif
 1165                         *(int *)data = pipe->pipe_buffer.size -
 1166                                         pipe->pipe_buffer.cnt;
 1167                 PIPE_UNLOCK(pipe);
 1168                 return (0);
 1169 
 1170         case TIOCSPGRP:
 1171         case FIOSETOWN:
 1172                 return fsetown(p, &pipe->pipe_pgid, cmd, data);
 1173 
 1174         case TIOCGPGRP:
 1175         case FIOGETOWN:
 1176                 return fgetown(p, pipe->pipe_pgid, cmd, data);
 1177 
 1178         }
 1179         return (EPASSTHROUGH);
 1180 }
 1181 
 1182 int
 1183 pipe_poll(fp, events, td)
 1184         struct file *fp;
 1185         int events;
 1186         struct proc *td;
 1187 {
 1188         struct pipe *rpipe = (struct pipe *)fp->f_data;
 1189         struct pipe *wpipe;
 1190         int eof = 0;
 1191         int revents = 0;
 1192 
 1193 retry:
 1194         PIPE_LOCK(rpipe);
 1195         wpipe = rpipe->pipe_peer;
 1196         if (wpipe != NULL && simple_lock_try(&wpipe->pipe_slock) == 0) {
 1197                 /* Deal with race for peer */
 1198                 PIPE_UNLOCK(rpipe);
 1199                 goto retry;
 1200         }
 1201 
 1202         if (events & (POLLIN | POLLRDNORM))
 1203                 if ((rpipe->pipe_buffer.cnt > 0) ||
 1204 #ifndef PIPE_NODIRECT
 1205                     (rpipe->pipe_state & PIPE_DIRECTR) ||
 1206 #endif
 1207                     (rpipe->pipe_state & PIPE_EOF))
 1208                         revents |= events & (POLLIN | POLLRDNORM);
 1209 
 1210         eof |= (rpipe->pipe_state & PIPE_EOF);
 1211         PIPE_UNLOCK(rpipe);
 1212 
 1213         if (wpipe == NULL)
 1214                 revents |= events & (POLLOUT | POLLWRNORM);
 1215         else {
 1216                 if (events & (POLLOUT | POLLWRNORM))
 1217                         if ((wpipe->pipe_state & PIPE_EOF) || (
 1218 #ifndef PIPE_NODIRECT
 1219                              (wpipe->pipe_state & PIPE_DIRECTW) == 0 &&
 1220 #endif
 1221                              (wpipe->pipe_buffer.size - wpipe->pipe_buffer.cnt) >= PIPE_BUF))
 1222                                 revents |= events & (POLLOUT | POLLWRNORM);
 1223 
 1224                 eof |= (wpipe->pipe_state & PIPE_EOF);
 1225                 PIPE_UNLOCK(wpipe);
 1226         }
 1227 
 1228         if (wpipe == NULL || eof)
 1229                 revents |= POLLHUP;
 1230 
 1231         if (revents == 0) {
 1232                 if (events & (POLLIN | POLLRDNORM))
 1233                         selrecord(td, &rpipe->pipe_sel);
 1234 
 1235                 if (events & (POLLOUT | POLLWRNORM))
 1236                         selrecord(td, &wpipe->pipe_sel);
 1237         }
 1238 
 1239         return (revents);
 1240 }
 1241 
 1242 static int
 1243 pipe_stat(fp, ub, td)
 1244         struct file *fp;
 1245         struct stat *ub;
 1246         struct proc *td;
 1247 {
 1248         struct pipe *pipe = (struct pipe *)fp->f_data;
 1249 
 1250         memset((caddr_t)ub, 0, sizeof(*ub));
 1251         ub->st_mode = S_IFIFO | S_IRUSR | S_IWUSR;
 1252         ub->st_blksize = pipe->pipe_buffer.size;
 1253         if (ub->st_blksize == 0 && pipe->pipe_peer)
 1254                 ub->st_blksize = pipe->pipe_peer->pipe_buffer.size;
 1255         ub->st_size = pipe->pipe_buffer.cnt;
 1256         ub->st_blocks = (ub->st_size) ? 1 : 0;
 1257         TIMEVAL_TO_TIMESPEC(&pipe->pipe_atime, &ub->st_atimespec);
 1258         TIMEVAL_TO_TIMESPEC(&pipe->pipe_mtime, &ub->st_mtimespec);
 1259         TIMEVAL_TO_TIMESPEC(&pipe->pipe_ctime, &ub->st_ctimespec);
 1260         ub->st_uid = fp->f_cred->cr_uid;
 1261         ub->st_gid = fp->f_cred->cr_gid;
 1262         /*
 1263          * Left as 0: st_dev, st_ino, st_nlink, st_rdev, st_flags, st_gen.
 1264          * XXX (st_dev, st_ino) should be unique.
 1265          */
 1266         return (0);
 1267 }
 1268 
 1269 /* ARGSUSED */
 1270 static int
 1271 pipe_close(fp, td)
 1272         struct file *fp;
 1273         struct proc *td;
 1274 {
 1275         struct pipe *pipe = (struct pipe *)fp->f_data;
 1276 
 1277         fp->f_data = NULL;
 1278         pipeclose(fp, pipe);
 1279         return (0);
 1280 }
 1281 
 1282 static void
 1283 pipe_free_kmem(pipe)
 1284         struct pipe *pipe;
 1285 {
 1286 
 1287         if (pipe->pipe_buffer.buffer != NULL) {
 1288                 if (pipe->pipe_buffer.size > PIPE_SIZE)
 1289                         --nbigpipe;
 1290                 amountpipekva -= pipe->pipe_buffer.size;
 1291                 uvm_km_free(kernel_map,
 1292                         (vaddr_t)pipe->pipe_buffer.buffer,
 1293                         pipe->pipe_buffer.size);
 1294                 pipe->pipe_buffer.buffer = NULL;
 1295         }
 1296 #ifndef PIPE_NODIRECT
 1297         if (pipe->pipe_map.kva != 0) {
 1298                 pipe_loan_free(pipe);
 1299                 pipe->pipe_map.cnt = 0;
 1300                 pipe->pipe_map.kva = 0;
 1301                 pipe->pipe_map.pos = 0;
 1302                 pipe->pipe_map.npages = 0;
 1303         }
 1304 #endif /* !PIPE_NODIRECT */
 1305 }
 1306 
 1307 /*
 1308  * shutdown the pipe
 1309  */
 1310 static void
 1311 pipeclose(fp, pipe)
 1312         struct file *fp;
 1313         struct pipe *pipe;
 1314 {
 1315         struct pipe *ppipe;
 1316 
 1317         if (pipe == NULL)
 1318                 return;
 1319 
 1320 retry:
 1321         PIPE_LOCK(pipe);
 1322 
 1323         pipeselwakeup(pipe, pipe, POLL_HUP);
 1324 
 1325         /*
 1326          * If the other side is blocked, wake it up saying that
 1327          * we want to close it down.
 1328          */
 1329         pipe->pipe_state |= PIPE_EOF;
 1330         while (pipe->pipe_busy) {
 1331                 wakeup(pipe);
 1332                 pipe->pipe_state |= PIPE_WANTCLOSE;
 1333                 ltsleep(pipe, PSOCK, "pipecl", 0, &pipe->pipe_slock);
 1334         }
 1335 
 1336         /*
 1337          * Disconnect from peer
 1338          */
 1339         if ((ppipe = pipe->pipe_peer) != NULL) {
 1340                 /* Deal with race for peer */
 1341                 if (simple_lock_try(&ppipe->pipe_slock) == 0) {
 1342                         PIPE_UNLOCK(pipe);
 1343                         goto retry;
 1344                 }
 1345                 pipeselwakeup(ppipe, ppipe, POLL_HUP);
 1346 
 1347                 ppipe->pipe_state |= PIPE_EOF;
 1348                 wakeup(ppipe);
 1349                 ppipe->pipe_peer = NULL;
 1350                 PIPE_UNLOCK(ppipe);
 1351         }
 1352 
 1353         (void)lockmgr(&pipe->pipe_lock, LK_DRAIN | LK_INTERLOCK,
 1354                         &pipe->pipe_slock);
 1355 
 1356         /*
 1357          * free resources
 1358          */
 1359         pipe_free_kmem(pipe);
 1360         pool_put(&pipe_pool, pipe);
 1361 }
 1362 
 1363 static void
 1364 filt_pipedetach(struct knote *kn)
 1365 {
 1366         struct pipe *pipe = (struct pipe *)kn->kn_fp->f_data;
 1367 
 1368         switch(kn->kn_filter) {
 1369         case EVFILT_WRITE:
 1370                 /* need the peer structure, not our own */
 1371                 pipe = pipe->pipe_peer;
 1372                 /* XXXSMP: race for peer */
 1373 
 1374                 /* if reader end already closed, just return */
 1375                 if (pipe == NULL)
 1376                         return;
 1377 
 1378                 break;
 1379         default:
 1380                 /* nothing to do */
 1381                 break;
 1382         }
 1383 
 1384 #ifdef DIAGNOSTIC
 1385         if (kn->kn_hook != pipe)
 1386                 panic("filt_pipedetach: inconsistent knote");
 1387 #endif
 1388 
 1389         PIPE_LOCK(pipe);
 1390         SLIST_REMOVE(&pipe->pipe_sel.sel_klist, kn, knote, kn_selnext);
 1391         PIPE_UNLOCK(pipe);
 1392 }
 1393 
 1394 /*ARGSUSED*/
 1395 static int
 1396 filt_piperead(struct knote *kn, long hint)
 1397 {
 1398         struct pipe *rpipe = (struct pipe *)kn->kn_fp->f_data;
 1399         struct pipe *wpipe = rpipe->pipe_peer;
 1400 
 1401         if ((hint & NOTE_SUBMIT) == 0)
 1402                 PIPE_LOCK(rpipe);
 1403         kn->kn_data = rpipe->pipe_buffer.cnt;
 1404         if ((kn->kn_data == 0) && (rpipe->pipe_state & PIPE_DIRECTW))
 1405                 kn->kn_data = rpipe->pipe_map.cnt;
 1406 
 1407         /* XXXSMP: race for peer */
 1408         if ((rpipe->pipe_state & PIPE_EOF) ||
 1409             (wpipe == NULL) || (wpipe->pipe_state & PIPE_EOF)) {
 1410                 kn->kn_flags |= EV_EOF;
 1411                 if ((hint & NOTE_SUBMIT) == 0)
 1412                         PIPE_UNLOCK(rpipe);
 1413                 return (1);
 1414         }
 1415         if ((hint & NOTE_SUBMIT) == 0)
 1416                 PIPE_UNLOCK(rpipe);
 1417         return (kn->kn_data > 0);
 1418 }
 1419 
 1420 /*ARGSUSED*/
 1421 static int
 1422 filt_pipewrite(struct knote *kn, long hint)
 1423 {
 1424         struct pipe *rpipe = (struct pipe *)kn->kn_fp->f_data;
 1425         struct pipe *wpipe = rpipe->pipe_peer;
 1426 
 1427         if ((hint & NOTE_SUBMIT) == 0)
 1428                 PIPE_LOCK(rpipe);
 1429         /* XXXSMP: race for peer */
 1430         if ((wpipe == NULL) || (wpipe->pipe_state & PIPE_EOF)) {
 1431                 kn->kn_data = 0;
 1432                 kn->kn_flags |= EV_EOF;
 1433                 if ((hint & NOTE_SUBMIT) == 0)
 1434                         PIPE_UNLOCK(rpipe);
 1435                 return (1);
 1436         }
 1437         kn->kn_data = wpipe->pipe_buffer.size - wpipe->pipe_buffer.cnt;
 1438         if (wpipe->pipe_state & PIPE_DIRECTW)
 1439                 kn->kn_data = 0;
 1440 
 1441         if ((hint & NOTE_SUBMIT) == 0)
 1442                 PIPE_UNLOCK(rpipe);
 1443         return (kn->kn_data >= PIPE_BUF);
 1444 }
 1445 
 1446 static const struct filterops pipe_rfiltops =
 1447         { 1, NULL, filt_pipedetach, filt_piperead };
 1448 static const struct filterops pipe_wfiltops =
 1449         { 1, NULL, filt_pipedetach, filt_pipewrite };
 1450 
 1451 /*ARGSUSED*/
 1452 static int
 1453 pipe_kqfilter(struct file *fp, struct knote *kn)
 1454 {
 1455         struct pipe *pipe;
 1456 
 1457         pipe = (struct pipe *)kn->kn_fp->f_data;
 1458         switch (kn->kn_filter) {
 1459         case EVFILT_READ:
 1460                 kn->kn_fop = &pipe_rfiltops;
 1461                 break;
 1462         case EVFILT_WRITE:
 1463                 kn->kn_fop = &pipe_wfiltops;
 1464                 /* XXXSMP: race for peer */
 1465                 pipe = pipe->pipe_peer;
 1466                 if (pipe == NULL) {
 1467                         /* other end of pipe has been closed */
 1468                         return (EBADF);
 1469                 }
 1470                 break;
 1471         default:
 1472                 return (1);
 1473         }
 1474         kn->kn_hook = pipe;
 1475 
 1476         PIPE_LOCK(pipe);
 1477         SLIST_INSERT_HEAD(&pipe->pipe_sel.sel_klist, kn, kn_selnext);
 1478         PIPE_UNLOCK(pipe);
 1479         return (0);
 1480 }
 1481 
 1482 /*
 1483  * Handle pipe sysctls.
 1484  */
 1485 SYSCTL_SETUP(sysctl_kern_pipe_setup, "sysctl kern.pipe subtree setup")
 1486 {
 1487 
 1488         sysctl_createv(clog, 0, NULL, NULL,
 1489                        CTLFLAG_PERMANENT,
 1490                        CTLTYPE_NODE, "kern", NULL,
 1491                        NULL, 0, NULL, 0,
 1492                        CTL_KERN, CTL_EOL);
 1493         sysctl_createv(clog, 0, NULL, NULL,
 1494                        CTLFLAG_PERMANENT,
 1495                        CTLTYPE_NODE, "pipe",
 1496                        SYSCTL_DESCR("Pipe settings"),
 1497                        NULL, 0, NULL, 0,
 1498                        CTL_KERN, KERN_PIPE, CTL_EOL);
 1499 
 1500         sysctl_createv(clog, 0, NULL, NULL,
 1501                        CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
 1502                        CTLTYPE_INT, "maxkvasz",
 1503                        SYSCTL_DESCR("Maximum amount of kernel memory to be "
 1504                                     "used for pipes"),
 1505                        NULL, 0, &maxpipekva, 0,
 1506                        CTL_KERN, KERN_PIPE, KERN_PIPE_MAXKVASZ, CTL_EOL);
 1507         sysctl_createv(clog, 0, NULL, NULL,
 1508                        CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
 1509                        CTLTYPE_INT, "maxloankvasz",
 1510                        SYSCTL_DESCR("Limit for direct transfers via page loan"),
 1511                        NULL, 0, &limitpipekva, 0,
 1512                        CTL_KERN, KERN_PIPE, KERN_PIPE_LIMITKVA, CTL_EOL);
 1513         sysctl_createv(clog, 0, NULL, NULL,
 1514                        CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
 1515                        CTLTYPE_INT, "maxbigpipes",
 1516                        SYSCTL_DESCR("Maximum number of \"big\" pipes"),
 1517                        NULL, 0, &maxbigpipes, 0,
 1518                        CTL_KERN, KERN_PIPE, KERN_PIPE_MAXBIGPIPES, CTL_EOL);
 1519         sysctl_createv(clog, 0, NULL, NULL,
 1520                        CTLFLAG_PERMANENT,
 1521                        CTLTYPE_INT, "nbigpipes",
 1522                        SYSCTL_DESCR("Number of \"big\" pipes"),
 1523                        NULL, 0, &nbigpipe, 0,
 1524                        CTL_KERN, KERN_PIPE, KERN_PIPE_NBIGPIPES, CTL_EOL);
 1525         sysctl_createv(clog, 0, NULL, NULL,
 1526                        CTLFLAG_PERMANENT,
 1527                        CTLTYPE_INT, "kvasize",
 1528                        SYSCTL_DESCR("Amount of kernel memory consumed by pipe "
 1529                                     "buffers"),
 1530                        NULL, 0, &amountpipekva, 0,
 1531                        CTL_KERN, KERN_PIPE, KERN_PIPE_KVASIZE, CTL_EOL);
 1532 }

Cache object: 1fc336c870c8f94e69a9e6f1c176e364


[ source navigation ] [ diff markup ] [ identifier search ] [ freetext search ] [ file search ] [ list types ] [ track identifier ]


This page is part of the FreeBSD/Linux Linux Kernel Cross-Reference, and was automatically generated using a modified version of the LXR engine.