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-12-STABLE  -  FREEBSD-12-0  -  FREEBSD-11-STABLE  -  FREEBSD-11-2  -  FREEBSD-11-1  -  FREEBSD-11-0  -  FREEBSD-10-STABLE  -  FREEBSD-10-4  -  FREEBSD-10-3  -  FREEBSD-10-2  -  FREEBSD-10-1  -  FREEBSD-10-0  -  FREEBSD-9-STABLE  -  FREEBSD-9-3  -  FREEBSD-9-2  -  FREEBSD-9-1  -  FREEBSD-9-0  -  FREEBSD-8-STABLE  -  FREEBSD-8-4  -  FREEBSD-8-3  -  FREEBSD-8-2  -  FREEBSD-8-1  -  FREEBSD-8-0  -  FREEBSD-7-STABLE  -  FREEBSD-7-4  -  FREEBSD-7-3  -  FREEBSD-7-2  -  FREEBSD-7-1  -  FREEBSD-7-0  -  FREEBSD-6-STABLE  -  FREEBSD-6-4  -  FREEBSD-6-3  -  FREEBSD-6-2  -  FREEBSD-6-1  -  FREEBSD-6-0  -  FREEBSD-5-STABLE  -  FREEBSD-5-5  -  FREEBSD-5-4  -  FREEBSD-5-3  -  FREEBSD-5-2  -  FREEBSD-5-1  -  FREEBSD-5-0  -  FREEBSD-4-STABLE  -  FREEBSD-3-STABLE  -  FREEBSD22  -  linux-2.6  -  linux-2.4.22  -  MK83  -  MK84  -  PLAN9  -  DFBSD  -  NETBSD  -  NETBSD5  -  NETBSD4  -  NETBSD3  -  NETBSD20  -  OPENBSD  -  xnu-517  -  xnu-792  -  xnu-792.6.70  -  xnu-1228  -  xnu-1456.1.26  -  xnu-1699.24.8  -  xnu-2050.18.24  -  OPENSOLARIS  -  minix-3-1-1 
SearchContext: -  none  -  3  -  10 

    1 /*-
    2  * SPDX-License-Identifier: BSD-4-Clause
    3  *
    4  * Copyright (c) 1996 John S. Dyson
    5  * Copyright (c) 2012 Giovanni Trematerra
    6  * All rights reserved.
    7  *
    8  * Redistribution and use in source and binary forms, with or without
    9  * modification, are permitted provided that the following conditions
   10  * are met:
   11  * 1. Redistributions of source code must retain the above copyright
   12  *    notice immediately at the beginning of the file, without modification,
   13  *    this list of conditions, and the following disclaimer.
   14  * 2. Redistributions in binary form must reproduce the above copyright
   15  *    notice, this list of conditions and the following disclaimer in the
   16  *    documentation and/or other materials provided with the distribution.
   17  * 3. Absolutely no warranty of function or purpose is made by the author
   18  *    John S. Dyson.
   19  * 4. Modifications may be freely made to this file if the above conditions
   20  *    are met.
   21  */
   22 
   23 /*
   24  * This file contains a high-performance replacement for the socket-based
   25  * pipes scheme originally used in FreeBSD/4.4Lite.  It does not support
   26  * all features of sockets, but does do everything that pipes normally
   27  * do.
   28  */
   29 
   30 /*
   31  * This code has two modes of operation, a small write mode and a large
   32  * write mode.  The small write mode acts like conventional pipes with
   33  * a kernel buffer.  If the buffer is less than PIPE_MINDIRECT, then the
   34  * "normal" pipe buffering is done.  If the buffer is between PIPE_MINDIRECT
   35  * and PIPE_SIZE in size, the sending process pins the underlying pages in
   36  * memory, and the receiving process copies directly from these pinned pages
   37  * in the sending process.
   38  *
   39  * If the sending process receives a signal, it is possible that it will
   40  * go away, and certainly its address space can change, because control
   41  * is returned back to the user-mode side.  In that case, the pipe code
   42  * arranges to copy the buffer supplied by the user process, to a pageable
   43  * kernel buffer, and the receiving process will grab the data from the
   44  * pageable kernel buffer.  Since signals don't happen all that often,
   45  * the copy operation is normally eliminated.
   46  *
   47  * The constant PIPE_MINDIRECT is chosen to make sure that buffering will
   48  * happen for small transfers so that the system will not spend all of
   49  * its time context switching.
   50  *
   51  * In order to limit the resource use of pipes, two sysctls exist:
   52  *
   53  * kern.ipc.maxpipekva - This is a hard limit on the amount of pageable
   54  * address space available to us in pipe_map. This value is normally
   55  * autotuned, but may also be loader tuned.
   56  *
   57  * kern.ipc.pipekva - This read-only sysctl tracks the current amount of
   58  * memory in use by pipes.
   59  *
   60  * Based on how large pipekva is relative to maxpipekva, the following
   61  * will happen:
   62  *
   63  * 0% - 50%:
   64  *     New pipes are given 16K of memory backing, pipes may dynamically
   65  *     grow to as large as 64K where needed.
   66  * 50% - 75%:
   67  *     New pipes are given 4K (or PAGE_SIZE) of memory backing,
   68  *     existing pipes may NOT grow.
   69  * 75% - 100%:
   70  *     New pipes are given 4K (or PAGE_SIZE) of memory backing,
   71  *     existing pipes will be shrunk down to 4K whenever possible.
   72  *
   73  * Resizing may be disabled by setting kern.ipc.piperesizeallowed=0.  If
   74  * that is set,  the only resize that will occur is the 0 -> SMALL_PIPE_SIZE
   75  * resize which MUST occur for reverse-direction pipes when they are
   76  * first used.
   77  *
   78  * Additional information about the current state of pipes may be obtained
   79  * from kern.ipc.pipes, kern.ipc.pipefragretry, kern.ipc.pipeallocfail,
   80  * and kern.ipc.piperesizefail.
   81  *
   82  * Locking rules:  There are two locks present here:  A mutex, used via
   83  * PIPE_LOCK, and a flag, used via pipelock().  All locking is done via
   84  * the flag, as mutexes can not persist over uiomove.  The mutex
   85  * exists only to guard access to the flag, and is not in itself a
   86  * locking mechanism.  Also note that there is only a single mutex for
   87  * both directions of a pipe.
   88  *
   89  * As pipelock() may have to sleep before it can acquire the flag, it
   90  * is important to reread all data after a call to pipelock(); everything
   91  * in the structure may have changed.
   92  */
   93 
   94 #include <sys/cdefs.h>
   95 __FBSDID("$FreeBSD: head/sys/kern/sys_pipe.c 344278 2019-02-19 15:46:43Z markj $");
   96 
   97 #include <sys/param.h>
   98 #include <sys/systm.h>
   99 #include <sys/conf.h>
  100 #include <sys/fcntl.h>
  101 #include <sys/file.h>
  102 #include <sys/filedesc.h>
  103 #include <sys/filio.h>
  104 #include <sys/kernel.h>
  105 #include <sys/lock.h>
  106 #include <sys/mutex.h>
  107 #include <sys/ttycom.h>
  108 #include <sys/stat.h>
  109 #include <sys/malloc.h>
  110 #include <sys/poll.h>
  111 #include <sys/selinfo.h>
  112 #include <sys/signalvar.h>
  113 #include <sys/syscallsubr.h>
  114 #include <sys/sysctl.h>
  115 #include <sys/sysproto.h>
  116 #include <sys/pipe.h>
  117 #include <sys/proc.h>
  118 #include <sys/vnode.h>
  119 #include <sys/uio.h>
  120 #include <sys/user.h>
  121 #include <sys/event.h>
  122 
  123 #include <security/mac/mac_framework.h>
  124 
  125 #include <vm/vm.h>
  126 #include <vm/vm_param.h>
  127 #include <vm/vm_object.h>
  128 #include <vm/vm_kern.h>
  129 #include <vm/vm_extern.h>
  130 #include <vm/pmap.h>
  131 #include <vm/vm_map.h>
  132 #include <vm/vm_page.h>
  133 #include <vm/uma.h>
  134 
  135 /*
  136  * Use this define if you want to disable *fancy* VM things.  Expect an
  137  * approx 30% decrease in transfer rate.  This could be useful for
  138  * NetBSD or OpenBSD.
  139  */
  140 /* #define PIPE_NODIRECT */
  141 
  142 #define PIPE_PEER(pipe) \
  143         (((pipe)->pipe_state & PIPE_NAMED) ? (pipe) : ((pipe)->pipe_peer))
  144 
  145 /*
  146  * interfaces to the outside world
  147  */
  148 static fo_rdwr_t        pipe_read;
  149 static fo_rdwr_t        pipe_write;
  150 static fo_truncate_t    pipe_truncate;
  151 static fo_ioctl_t       pipe_ioctl;
  152 static fo_poll_t        pipe_poll;
  153 static fo_kqfilter_t    pipe_kqfilter;
  154 static fo_stat_t        pipe_stat;
  155 static fo_close_t       pipe_close;
  156 static fo_chmod_t       pipe_chmod;
  157 static fo_chown_t       pipe_chown;
  158 static fo_fill_kinfo_t  pipe_fill_kinfo;
  159 
  160 struct fileops pipeops = {
  161         .fo_read = pipe_read,
  162         .fo_write = pipe_write,
  163         .fo_truncate = pipe_truncate,
  164         .fo_ioctl = pipe_ioctl,
  165         .fo_poll = pipe_poll,
  166         .fo_kqfilter = pipe_kqfilter,
  167         .fo_stat = pipe_stat,
  168         .fo_close = pipe_close,
  169         .fo_chmod = pipe_chmod,
  170         .fo_chown = pipe_chown,
  171         .fo_sendfile = invfo_sendfile,
  172         .fo_fill_kinfo = pipe_fill_kinfo,
  173         .fo_flags = DFLAG_PASSABLE
  174 };
  175 
  176 static void     filt_pipedetach(struct knote *kn);
  177 static void     filt_pipedetach_notsup(struct knote *kn);
  178 static int      filt_pipenotsup(struct knote *kn, long hint);
  179 static int      filt_piperead(struct knote *kn, long hint);
  180 static int      filt_pipewrite(struct knote *kn, long hint);
  181 
  182 static struct filterops pipe_nfiltops = {
  183         .f_isfd = 1,
  184         .f_detach = filt_pipedetach_notsup,
  185         .f_event = filt_pipenotsup
  186 };
  187 static struct filterops pipe_rfiltops = {
  188         .f_isfd = 1,
  189         .f_detach = filt_pipedetach,
  190         .f_event = filt_piperead
  191 };
  192 static struct filterops pipe_wfiltops = {
  193         .f_isfd = 1,
  194         .f_detach = filt_pipedetach,
  195         .f_event = filt_pipewrite
  196 };
  197 
  198 /*
  199  * Default pipe buffer size(s), this can be kind-of large now because pipe
  200  * space is pageable.  The pipe code will try to maintain locality of
  201  * reference for performance reasons, so small amounts of outstanding I/O
  202  * will not wipe the cache.
  203  */
  204 #define MINPIPESIZE (PIPE_SIZE/3)
  205 #define MAXPIPESIZE (2*PIPE_SIZE/3)
  206 
  207 static long amountpipekva;
  208 static int pipefragretry;
  209 static int pipeallocfail;
  210 static int piperesizefail;
  211 static int piperesizeallowed = 1;
  212 
  213 SYSCTL_LONG(_kern_ipc, OID_AUTO, maxpipekva, CTLFLAG_RDTUN | CTLFLAG_NOFETCH,
  214            &maxpipekva, 0, "Pipe KVA limit");
  215 SYSCTL_LONG(_kern_ipc, OID_AUTO, pipekva, CTLFLAG_RD,
  216            &amountpipekva, 0, "Pipe KVA usage");
  217 SYSCTL_INT(_kern_ipc, OID_AUTO, pipefragretry, CTLFLAG_RD,
  218           &pipefragretry, 0, "Pipe allocation retries due to fragmentation");
  219 SYSCTL_INT(_kern_ipc, OID_AUTO, pipeallocfail, CTLFLAG_RD,
  220           &pipeallocfail, 0, "Pipe allocation failures");
  221 SYSCTL_INT(_kern_ipc, OID_AUTO, piperesizefail, CTLFLAG_RD,
  222           &piperesizefail, 0, "Pipe resize failures");
  223 SYSCTL_INT(_kern_ipc, OID_AUTO, piperesizeallowed, CTLFLAG_RW,
  224           &piperesizeallowed, 0, "Pipe resizing allowed");
  225 
  226 static void pipeinit(void *dummy __unused);
  227 static void pipeclose(struct pipe *cpipe);
  228 static void pipe_free_kmem(struct pipe *cpipe);
  229 static void pipe_create(struct pipe *pipe, int backing);
  230 static void pipe_paircreate(struct thread *td, struct pipepair **p_pp);
  231 static __inline int pipelock(struct pipe *cpipe, int catch);
  232 static __inline void pipeunlock(struct pipe *cpipe);
  233 #ifndef PIPE_NODIRECT
  234 static int pipe_build_write_buffer(struct pipe *wpipe, struct uio *uio);
  235 static void pipe_destroy_write_buffer(struct pipe *wpipe);
  236 static int pipe_direct_write(struct pipe *wpipe, struct uio *uio);
  237 static void pipe_clone_write_buffer(struct pipe *wpipe);
  238 #endif
  239 static int pipespace(struct pipe *cpipe, int size);
  240 static int pipespace_new(struct pipe *cpipe, int size);
  241 
  242 static int      pipe_zone_ctor(void *mem, int size, void *arg, int flags);
  243 static int      pipe_zone_init(void *mem, int size, int flags);
  244 static void     pipe_zone_fini(void *mem, int size);
  245 
  246 static uma_zone_t pipe_zone;
  247 static struct unrhdr64 pipeino_unr;
  248 static dev_t pipedev_ino;
  249 
  250 SYSINIT(vfs, SI_SUB_VFS, SI_ORDER_ANY, pipeinit, NULL);
  251 
  252 static void
  253 pipeinit(void *dummy __unused)
  254 {
  255 
  256         pipe_zone = uma_zcreate("pipe", sizeof(struct pipepair),
  257             pipe_zone_ctor, NULL, pipe_zone_init, pipe_zone_fini,
  258             UMA_ALIGN_PTR, 0);
  259         KASSERT(pipe_zone != NULL, ("pipe_zone not initialized"));
  260         new_unrhdr64(&pipeino_unr, 1);
  261         pipedev_ino = devfs_alloc_cdp_inode();
  262         KASSERT(pipedev_ino > 0, ("pipe dev inode not initialized"));
  263 }
  264 
  265 static int
  266 pipe_zone_ctor(void *mem, int size, void *arg, int flags)
  267 {
  268         struct pipepair *pp;
  269         struct pipe *rpipe, *wpipe;
  270 
  271         KASSERT(size == sizeof(*pp), ("pipe_zone_ctor: wrong size"));
  272 
  273         pp = (struct pipepair *)mem;
  274 
  275         /*
  276          * We zero both pipe endpoints to make sure all the kmem pointers
  277          * are NULL, flag fields are zero'd, etc.  We timestamp both
  278          * endpoints with the same time.
  279          */
  280         rpipe = &pp->pp_rpipe;
  281         bzero(rpipe, sizeof(*rpipe));
  282         vfs_timestamp(&rpipe->pipe_ctime);
  283         rpipe->pipe_atime = rpipe->pipe_mtime = rpipe->pipe_ctime;
  284 
  285         wpipe = &pp->pp_wpipe;
  286         bzero(wpipe, sizeof(*wpipe));
  287         wpipe->pipe_ctime = rpipe->pipe_ctime;
  288         wpipe->pipe_atime = wpipe->pipe_mtime = rpipe->pipe_ctime;
  289 
  290         rpipe->pipe_peer = wpipe;
  291         rpipe->pipe_pair = pp;
  292         wpipe->pipe_peer = rpipe;
  293         wpipe->pipe_pair = pp;
  294 
  295         /*
  296          * Mark both endpoints as present; they will later get free'd
  297          * one at a time.  When both are free'd, then the whole pair
  298          * is released.
  299          */
  300         rpipe->pipe_present = PIPE_ACTIVE;
  301         wpipe->pipe_present = PIPE_ACTIVE;
  302 
  303         /*
  304          * Eventually, the MAC Framework may initialize the label
  305          * in ctor or init, but for now we do it elswhere to avoid
  306          * blocking in ctor or init.
  307          */
  308         pp->pp_label = NULL;
  309 
  310         return (0);
  311 }
  312 
  313 static int
  314 pipe_zone_init(void *mem, int size, int flags)
  315 {
  316         struct pipepair *pp;
  317 
  318         KASSERT(size == sizeof(*pp), ("pipe_zone_init: wrong size"));
  319 
  320         pp = (struct pipepair *)mem;
  321 
  322         mtx_init(&pp->pp_mtx, "pipe mutex", NULL, MTX_DEF | MTX_NEW);
  323         return (0);
  324 }
  325 
  326 static void
  327 pipe_zone_fini(void *mem, int size)
  328 {
  329         struct pipepair *pp;
  330 
  331         KASSERT(size == sizeof(*pp), ("pipe_zone_fini: wrong size"));
  332 
  333         pp = (struct pipepair *)mem;
  334 
  335         mtx_destroy(&pp->pp_mtx);
  336 }
  337 
  338 static void
  339 pipe_paircreate(struct thread *td, struct pipepair **p_pp)
  340 {
  341         struct pipepair *pp;
  342         struct pipe *rpipe, *wpipe;
  343 
  344         *p_pp = pp = uma_zalloc(pipe_zone, M_WAITOK);
  345 #ifdef MAC
  346         /*
  347          * The MAC label is shared between the connected endpoints.  As a
  348          * result mac_pipe_init() and mac_pipe_create() are called once
  349          * for the pair, and not on the endpoints.
  350          */
  351         mac_pipe_init(pp);
  352         mac_pipe_create(td->td_ucred, pp);
  353 #endif
  354         rpipe = &pp->pp_rpipe;
  355         wpipe = &pp->pp_wpipe;
  356 
  357         knlist_init_mtx(&rpipe->pipe_sel.si_note, PIPE_MTX(rpipe));
  358         knlist_init_mtx(&wpipe->pipe_sel.si_note, PIPE_MTX(wpipe));
  359 
  360         /* Only the forward direction pipe is backed by default */
  361         pipe_create(rpipe, 1);
  362         pipe_create(wpipe, 0);
  363 
  364         rpipe->pipe_state |= PIPE_DIRECTOK;
  365         wpipe->pipe_state |= PIPE_DIRECTOK;
  366 }
  367 
  368 void
  369 pipe_named_ctor(struct pipe **ppipe, struct thread *td)
  370 {
  371         struct pipepair *pp;
  372 
  373         pipe_paircreate(td, &pp);
  374         pp->pp_rpipe.pipe_state |= PIPE_NAMED;
  375         *ppipe = &pp->pp_rpipe;
  376 }
  377 
  378 void
  379 pipe_dtor(struct pipe *dpipe)
  380 {
  381         struct pipe *peer;
  382 
  383         peer = (dpipe->pipe_state & PIPE_NAMED) != 0 ? dpipe->pipe_peer : NULL;
  384         funsetown(&dpipe->pipe_sigio);
  385         pipeclose(dpipe);
  386         if (peer != NULL) {
  387                 funsetown(&peer->pipe_sigio);
  388                 pipeclose(peer);
  389         }
  390 }
  391 
  392 /*
  393  * The pipe system call for the DTYPE_PIPE type of pipes.  If we fail, let
  394  * the zone pick up the pieces via pipeclose().
  395  */
  396 int
  397 kern_pipe(struct thread *td, int fildes[2], int flags, struct filecaps *fcaps1,
  398     struct filecaps *fcaps2)
  399 {
  400         struct file *rf, *wf;
  401         struct pipe *rpipe, *wpipe;
  402         struct pipepair *pp;
  403         int fd, fflags, error;
  404 
  405         pipe_paircreate(td, &pp);
  406         rpipe = &pp->pp_rpipe;
  407         wpipe = &pp->pp_wpipe;
  408         error = falloc_caps(td, &rf, &fd, flags, fcaps1);
  409         if (error) {
  410                 pipeclose(rpipe);
  411                 pipeclose(wpipe);
  412                 return (error);
  413         }
  414         /* An extra reference on `rf' has been held for us by falloc_caps(). */
  415         fildes[0] = fd;
  416 
  417         fflags = FREAD | FWRITE;
  418         if ((flags & O_NONBLOCK) != 0)
  419                 fflags |= FNONBLOCK;
  420 
  421         /*
  422          * Warning: once we've gotten past allocation of the fd for the
  423          * read-side, we can only drop the read side via fdrop() in order
  424          * to avoid races against processes which manage to dup() the read
  425          * side while we are blocked trying to allocate the write side.
  426          */
  427         finit(rf, fflags, DTYPE_PIPE, rpipe, &pipeops);
  428         error = falloc_caps(td, &wf, &fd, flags, fcaps2);
  429         if (error) {
  430                 fdclose(td, rf, fildes[0]);
  431                 fdrop(rf, td);
  432                 /* rpipe has been closed by fdrop(). */
  433                 pipeclose(wpipe);
  434                 return (error);
  435         }
  436         /* An extra reference on `wf' has been held for us by falloc_caps(). */
  437         finit(wf, fflags, DTYPE_PIPE, wpipe, &pipeops);
  438         fdrop(wf, td);
  439         fildes[1] = fd;
  440         fdrop(rf, td);
  441 
  442         return (0);
  443 }
  444 
  445 #ifdef COMPAT_FREEBSD10
  446 /* ARGSUSED */
  447 int
  448 freebsd10_pipe(struct thread *td, struct freebsd10_pipe_args *uap __unused)
  449 {
  450         int error;
  451         int fildes[2];
  452 
  453         error = kern_pipe(td, fildes, 0, NULL, NULL);
  454         if (error)
  455                 return (error);
  456 
  457         td->td_retval[0] = fildes[0];
  458         td->td_retval[1] = fildes[1];
  459 
  460         return (0);
  461 }
  462 #endif
  463 
  464 int
  465 sys_pipe2(struct thread *td, struct pipe2_args *uap)
  466 {
  467         int error, fildes[2];
  468 
  469         if (uap->flags & ~(O_CLOEXEC | O_NONBLOCK))
  470                 return (EINVAL);
  471         error = kern_pipe(td, fildes, uap->flags, NULL, NULL);
  472         if (error)
  473                 return (error);
  474         error = copyout(fildes, uap->fildes, 2 * sizeof(int));
  475         if (error) {
  476                 (void)kern_close(td, fildes[0]);
  477                 (void)kern_close(td, fildes[1]);
  478         }
  479         return (error);
  480 }
  481 
  482 /*
  483  * Allocate kva for pipe circular buffer, the space is pageable
  484  * This routine will 'realloc' the size of a pipe safely, if it fails
  485  * it will retain the old buffer.
  486  * If it fails it will return ENOMEM.
  487  */
  488 static int
  489 pipespace_new(struct pipe *cpipe, int size)
  490 {
  491         caddr_t buffer;
  492         int error, cnt, firstseg;
  493         static int curfail = 0;
  494         static struct timeval lastfail;
  495 
  496         KASSERT(!mtx_owned(PIPE_MTX(cpipe)), ("pipespace: pipe mutex locked"));
  497         KASSERT(!(cpipe->pipe_state & PIPE_DIRECTW),
  498                 ("pipespace: resize of direct writes not allowed"));
  499 retry:
  500         cnt = cpipe->pipe_buffer.cnt;
  501         if (cnt > size)
  502                 size = cnt;
  503 
  504         size = round_page(size);
  505         buffer = (caddr_t) vm_map_min(pipe_map);
  506 
  507         error = vm_map_find(pipe_map, NULL, 0, (vm_offset_t *)&buffer, size, 0,
  508             VMFS_ANY_SPACE, VM_PROT_RW, VM_PROT_RW, 0);
  509         if (error != KERN_SUCCESS) {
  510                 if ((cpipe->pipe_buffer.buffer == NULL) &&
  511                         (size > SMALL_PIPE_SIZE)) {
  512                         size = SMALL_PIPE_SIZE;
  513                         pipefragretry++;
  514                         goto retry;
  515                 }
  516                 if (cpipe->pipe_buffer.buffer == NULL) {
  517                         pipeallocfail++;
  518                         if (ppsratecheck(&lastfail, &curfail, 1))
  519                                 printf("kern.ipc.maxpipekva exceeded; see tuning(7)\n");
  520                 } else {
  521                         piperesizefail++;
  522                 }
  523                 return (ENOMEM);
  524         }
  525 
  526         /* copy data, then free old resources if we're resizing */
  527         if (cnt > 0) {
  528                 if (cpipe->pipe_buffer.in <= cpipe->pipe_buffer.out) {
  529                         firstseg = cpipe->pipe_buffer.size - cpipe->pipe_buffer.out;
  530                         bcopy(&cpipe->pipe_buffer.buffer[cpipe->pipe_buffer.out],
  531                                 buffer, firstseg);
  532                         if ((cnt - firstseg) > 0)
  533                                 bcopy(cpipe->pipe_buffer.buffer, &buffer[firstseg],
  534                                         cpipe->pipe_buffer.in);
  535                 } else {
  536                         bcopy(&cpipe->pipe_buffer.buffer[cpipe->pipe_buffer.out],
  537                                 buffer, cnt);
  538                 }
  539         }
  540         pipe_free_kmem(cpipe);
  541         cpipe->pipe_buffer.buffer = buffer;
  542         cpipe->pipe_buffer.size = size;
  543         cpipe->pipe_buffer.in = cnt;
  544         cpipe->pipe_buffer.out = 0;
  545         cpipe->pipe_buffer.cnt = cnt;
  546         atomic_add_long(&amountpipekva, cpipe->pipe_buffer.size);
  547         return (0);
  548 }
  549 
  550 /*
  551  * Wrapper for pipespace_new() that performs locking assertions.
  552  */
  553 static int
  554 pipespace(struct pipe *cpipe, int size)
  555 {
  556 
  557         KASSERT(cpipe->pipe_state & PIPE_LOCKFL,
  558                 ("Unlocked pipe passed to pipespace"));
  559         return (pipespace_new(cpipe, size));
  560 }
  561 
  562 /*
  563  * lock a pipe for I/O, blocking other access
  564  */
  565 static __inline int
  566 pipelock(struct pipe *cpipe, int catch)
  567 {
  568         int error;
  569 
  570         PIPE_LOCK_ASSERT(cpipe, MA_OWNED);
  571         while (cpipe->pipe_state & PIPE_LOCKFL) {
  572                 cpipe->pipe_state |= PIPE_LWANT;
  573                 error = msleep(cpipe, PIPE_MTX(cpipe),
  574                     catch ? (PRIBIO | PCATCH) : PRIBIO,
  575                     "pipelk", 0);
  576                 if (error != 0)
  577                         return (error);
  578         }
  579         cpipe->pipe_state |= PIPE_LOCKFL;
  580         return (0);
  581 }
  582 
  583 /*
  584  * unlock a pipe I/O lock
  585  */
  586 static __inline void
  587 pipeunlock(struct pipe *cpipe)
  588 {
  589 
  590         PIPE_LOCK_ASSERT(cpipe, MA_OWNED);
  591         KASSERT(cpipe->pipe_state & PIPE_LOCKFL,
  592                 ("Unlocked pipe passed to pipeunlock"));
  593         cpipe->pipe_state &= ~PIPE_LOCKFL;
  594         if (cpipe->pipe_state & PIPE_LWANT) {
  595                 cpipe->pipe_state &= ~PIPE_LWANT;
  596                 wakeup(cpipe);
  597         }
  598 }
  599 
  600 void
  601 pipeselwakeup(struct pipe *cpipe)
  602 {
  603 
  604         PIPE_LOCK_ASSERT(cpipe, MA_OWNED);
  605         if (cpipe->pipe_state & PIPE_SEL) {
  606                 selwakeuppri(&cpipe->pipe_sel, PSOCK);
  607                 if (!SEL_WAITING(&cpipe->pipe_sel))
  608                         cpipe->pipe_state &= ~PIPE_SEL;
  609         }
  610         if ((cpipe->pipe_state & PIPE_ASYNC) && cpipe->pipe_sigio)
  611                 pgsigio(&cpipe->pipe_sigio, SIGIO, 0);
  612         KNOTE_LOCKED(&cpipe->pipe_sel.si_note, 0);
  613 }
  614 
  615 /*
  616  * Initialize and allocate VM and memory for pipe.  The structure
  617  * will start out zero'd from the ctor, so we just manage the kmem.
  618  */
  619 static void
  620 pipe_create(struct pipe *pipe, int backing)
  621 {
  622 
  623         if (backing) {
  624                 /*
  625                  * Note that these functions can fail if pipe map is exhausted
  626                  * (as a result of too many pipes created), but we ignore the
  627                  * error as it is not fatal and could be provoked by
  628                  * unprivileged users. The only consequence is worse performance
  629                  * with given pipe.
  630                  */
  631                 if (amountpipekva > maxpipekva / 2)
  632                         (void)pipespace_new(pipe, SMALL_PIPE_SIZE);
  633                 else
  634                         (void)pipespace_new(pipe, PIPE_SIZE);
  635         }
  636 
  637         pipe->pipe_ino = alloc_unr64(&pipeino_unr);
  638 }
  639 
  640 /* ARGSUSED */
  641 static int
  642 pipe_read(struct file *fp, struct uio *uio, struct ucred *active_cred,
  643     int flags, struct thread *td)
  644 {
  645         struct pipe *rpipe;
  646         int error;
  647         int nread = 0;
  648         int size;
  649 
  650         rpipe = fp->f_data;
  651         PIPE_LOCK(rpipe);
  652         ++rpipe->pipe_busy;
  653         error = pipelock(rpipe, 1);
  654         if (error)
  655                 goto unlocked_error;
  656 
  657 #ifdef MAC
  658         error = mac_pipe_check_read(active_cred, rpipe->pipe_pair);
  659         if (error)
  660                 goto locked_error;
  661 #endif
  662         if (amountpipekva > (3 * maxpipekva) / 4) {
  663                 if (!(rpipe->pipe_state & PIPE_DIRECTW) &&
  664                         (rpipe->pipe_buffer.size > SMALL_PIPE_SIZE) &&
  665                         (rpipe->pipe_buffer.cnt <= SMALL_PIPE_SIZE) &&
  666                         (piperesizeallowed == 1)) {
  667                         PIPE_UNLOCK(rpipe);
  668                         pipespace(rpipe, SMALL_PIPE_SIZE);
  669                         PIPE_LOCK(rpipe);
  670                 }
  671         }
  672 
  673         while (uio->uio_resid) {
  674                 /*
  675                  * normal pipe buffer receive
  676                  */
  677                 if (rpipe->pipe_buffer.cnt > 0) {
  678                         size = rpipe->pipe_buffer.size - rpipe->pipe_buffer.out;
  679                         if (size > rpipe->pipe_buffer.cnt)
  680                                 size = rpipe->pipe_buffer.cnt;
  681                         if (size > uio->uio_resid)
  682                                 size = uio->uio_resid;
  683 
  684                         PIPE_UNLOCK(rpipe);
  685                         error = uiomove(
  686                             &rpipe->pipe_buffer.buffer[rpipe->pipe_buffer.out],
  687                             size, uio);
  688                         PIPE_LOCK(rpipe);
  689                         if (error)
  690                                 break;
  691 
  692                         rpipe->pipe_buffer.out += size;
  693                         if (rpipe->pipe_buffer.out >= rpipe->pipe_buffer.size)
  694                                 rpipe->pipe_buffer.out = 0;
  695 
  696                         rpipe->pipe_buffer.cnt -= size;
  697 
  698                         /*
  699                          * If there is no more to read in the pipe, reset
  700                          * its pointers to the beginning.  This improves
  701                          * cache hit stats.
  702                          */
  703                         if (rpipe->pipe_buffer.cnt == 0) {
  704                                 rpipe->pipe_buffer.in = 0;
  705                                 rpipe->pipe_buffer.out = 0;
  706                         }
  707                         nread += size;
  708 #ifndef PIPE_NODIRECT
  709                 /*
  710                  * Direct copy, bypassing a kernel buffer.
  711                  */
  712                 } else if ((size = rpipe->pipe_map.cnt) &&
  713                            (rpipe->pipe_state & PIPE_DIRECTW)) {
  714                         if (size > uio->uio_resid)
  715                                 size = (u_int) uio->uio_resid;
  716 
  717                         PIPE_UNLOCK(rpipe);
  718                         error = uiomove_fromphys(rpipe->pipe_map.ms,
  719                             rpipe->pipe_map.pos, size, uio);
  720                         PIPE_LOCK(rpipe);
  721                         if (error)
  722                                 break;
  723                         nread += size;
  724                         rpipe->pipe_map.pos += size;
  725                         rpipe->pipe_map.cnt -= size;
  726                         if (rpipe->pipe_map.cnt == 0) {
  727                                 rpipe->pipe_state &= ~(PIPE_DIRECTW|PIPE_WANTW);
  728                                 wakeup(rpipe);
  729                         }
  730 #endif
  731                 } else {
  732                         /*
  733                          * detect EOF condition
  734                          * read returns 0 on EOF, no need to set error
  735                          */
  736                         if (rpipe->pipe_state & PIPE_EOF)
  737                                 break;
  738 
  739                         /*
  740                          * If the "write-side" has been blocked, wake it up now.
  741                          */
  742                         if (rpipe->pipe_state & PIPE_WANTW) {
  743                                 rpipe->pipe_state &= ~PIPE_WANTW;
  744                                 wakeup(rpipe);
  745                         }
  746 
  747                         /*
  748                          * Break if some data was read.
  749                          */
  750                         if (nread > 0)
  751                                 break;
  752 
  753                         /*
  754                          * Unlock the pipe buffer for our remaining processing.
  755                          * We will either break out with an error or we will
  756                          * sleep and relock to loop.
  757                          */
  758                         pipeunlock(rpipe);
  759 
  760                         /*
  761                          * Handle non-blocking mode operation or
  762                          * wait for more data.
  763                          */
  764                         if (fp->f_flag & FNONBLOCK) {
  765                                 error = EAGAIN;
  766                         } else {
  767                                 rpipe->pipe_state |= PIPE_WANTR;
  768                                 if ((error = msleep(rpipe, PIPE_MTX(rpipe),
  769                                     PRIBIO | PCATCH,
  770                                     "piperd", 0)) == 0)
  771                                         error = pipelock(rpipe, 1);
  772                         }
  773                         if (error)
  774                                 goto unlocked_error;
  775                 }
  776         }
  777 #ifdef MAC
  778 locked_error:
  779 #endif
  780         pipeunlock(rpipe);
  781 
  782         /* XXX: should probably do this before getting any locks. */
  783         if (error == 0)
  784                 vfs_timestamp(&rpipe->pipe_atime);
  785 unlocked_error:
  786         --rpipe->pipe_busy;
  787 
  788         /*
  789          * PIPE_WANT processing only makes sense if pipe_busy is 0.
  790          */
  791         if ((rpipe->pipe_busy == 0) && (rpipe->pipe_state & PIPE_WANT)) {
  792                 rpipe->pipe_state &= ~(PIPE_WANT|PIPE_WANTW);
  793                 wakeup(rpipe);
  794         } else if (rpipe->pipe_buffer.cnt < MINPIPESIZE) {
  795                 /*
  796                  * Handle write blocking hysteresis.
  797                  */
  798                 if (rpipe->pipe_state & PIPE_WANTW) {
  799                         rpipe->pipe_state &= ~PIPE_WANTW;
  800                         wakeup(rpipe);
  801                 }
  802         }
  803 
  804         if ((rpipe->pipe_buffer.size - rpipe->pipe_buffer.cnt) >= PIPE_BUF)
  805                 pipeselwakeup(rpipe);
  806 
  807         PIPE_UNLOCK(rpipe);
  808         return (error);
  809 }
  810 
  811 #ifndef PIPE_NODIRECT
  812 /*
  813  * Map the sending processes' buffer into kernel space and wire it.
  814  * This is similar to a physical write operation.
  815  */
  816 static int
  817 pipe_build_write_buffer(struct pipe *wpipe, struct uio *uio)
  818 {
  819         u_int size;
  820         int i;
  821 
  822         PIPE_LOCK_ASSERT(wpipe, MA_NOTOWNED);
  823         KASSERT(wpipe->pipe_state & PIPE_DIRECTW,
  824                 ("Clone attempt on non-direct write pipe!"));
  825 
  826         if (uio->uio_iov->iov_len > wpipe->pipe_buffer.size)
  827                 size = wpipe->pipe_buffer.size;
  828         else
  829                 size = uio->uio_iov->iov_len;
  830 
  831         if ((i = vm_fault_quick_hold_pages(&curproc->p_vmspace->vm_map,
  832             (vm_offset_t)uio->uio_iov->iov_base, size, VM_PROT_READ,
  833             wpipe->pipe_map.ms, PIPENPAGES)) < 0)
  834                 return (EFAULT);
  835 
  836 /*
  837  * set up the control block
  838  */
  839         wpipe->pipe_map.npages = i;
  840         wpipe->pipe_map.pos =
  841             ((vm_offset_t) uio->uio_iov->iov_base) & PAGE_MASK;
  842         wpipe->pipe_map.cnt = size;
  843 
  844 /*
  845  * and update the uio data
  846  */
  847 
  848         uio->uio_iov->iov_len -= size;
  849         uio->uio_iov->iov_base = (char *)uio->uio_iov->iov_base + size;
  850         if (uio->uio_iov->iov_len == 0)
  851                 uio->uio_iov++;
  852         uio->uio_resid -= size;
  853         uio->uio_offset += size;
  854         return (0);
  855 }
  856 
  857 /*
  858  * unmap and unwire the process buffer
  859  */
  860 static void
  861 pipe_destroy_write_buffer(struct pipe *wpipe)
  862 {
  863 
  864         PIPE_LOCK_ASSERT(wpipe, MA_OWNED);
  865         vm_page_unhold_pages(wpipe->pipe_map.ms, wpipe->pipe_map.npages);
  866         wpipe->pipe_map.npages = 0;
  867 }
  868 
  869 /*
  870  * In the case of a signal, the writing process might go away.  This
  871  * code copies the data into the circular buffer so that the source
  872  * pages can be freed without loss of data.
  873  */
  874 static void
  875 pipe_clone_write_buffer(struct pipe *wpipe)
  876 {
  877         struct uio uio;
  878         struct iovec iov;
  879         int size;
  880         int pos;
  881 
  882         PIPE_LOCK_ASSERT(wpipe, MA_OWNED);
  883         size = wpipe->pipe_map.cnt;
  884         pos = wpipe->pipe_map.pos;
  885 
  886         wpipe->pipe_buffer.in = size;
  887         wpipe->pipe_buffer.out = 0;
  888         wpipe->pipe_buffer.cnt = size;
  889         wpipe->pipe_state &= ~PIPE_DIRECTW;
  890 
  891         PIPE_UNLOCK(wpipe);
  892         iov.iov_base = wpipe->pipe_buffer.buffer;
  893         iov.iov_len = size;
  894         uio.uio_iov = &iov;
  895         uio.uio_iovcnt = 1;
  896         uio.uio_offset = 0;
  897         uio.uio_resid = size;
  898         uio.uio_segflg = UIO_SYSSPACE;
  899         uio.uio_rw = UIO_READ;
  900         uio.uio_td = curthread;
  901         uiomove_fromphys(wpipe->pipe_map.ms, pos, size, &uio);
  902         PIPE_LOCK(wpipe);
  903         pipe_destroy_write_buffer(wpipe);
  904 }
  905 
  906 /*
  907  * This implements the pipe buffer write mechanism.  Note that only
  908  * a direct write OR a normal pipe write can be pending at any given time.
  909  * If there are any characters in the pipe buffer, the direct write will
  910  * be deferred until the receiving process grabs all of the bytes from
  911  * the pipe buffer.  Then the direct mapping write is set-up.
  912  */
  913 static int
  914 pipe_direct_write(struct pipe *wpipe, struct uio *uio)
  915 {
  916         int error;
  917 
  918 retry:
  919         PIPE_LOCK_ASSERT(wpipe, MA_OWNED);
  920         error = pipelock(wpipe, 1);
  921         if (error != 0)
  922                 goto error1;
  923         if ((wpipe->pipe_state & PIPE_EOF) != 0) {
  924                 error = EPIPE;
  925                 pipeunlock(wpipe);
  926                 goto error1;
  927         }
  928         while (wpipe->pipe_state & PIPE_DIRECTW) {
  929                 if (wpipe->pipe_state & PIPE_WANTR) {
  930                         wpipe->pipe_state &= ~PIPE_WANTR;
  931                         wakeup(wpipe);
  932                 }
  933                 pipeselwakeup(wpipe);
  934                 wpipe->pipe_state |= PIPE_WANTW;
  935                 pipeunlock(wpipe);
  936                 error = msleep(wpipe, PIPE_MTX(wpipe),
  937                     PRIBIO | PCATCH, "pipdww", 0);
  938                 if (error)
  939                         goto error1;
  940                 else
  941                         goto retry;
  942         }
  943         wpipe->pipe_map.cnt = 0;        /* transfer not ready yet */
  944         if (wpipe->pipe_buffer.cnt > 0) {
  945                 if (wpipe->pipe_state & PIPE_WANTR) {
  946                         wpipe->pipe_state &= ~PIPE_WANTR;
  947                         wakeup(wpipe);
  948                 }
  949                 pipeselwakeup(wpipe);
  950                 wpipe->pipe_state |= PIPE_WANTW;
  951                 pipeunlock(wpipe);
  952                 error = msleep(wpipe, PIPE_MTX(wpipe),
  953                     PRIBIO | PCATCH, "pipdwc", 0);
  954                 if (error)
  955                         goto error1;
  956                 else
  957                         goto retry;
  958         }
  959 
  960         wpipe->pipe_state |= PIPE_DIRECTW;
  961 
  962         PIPE_UNLOCK(wpipe);
  963         error = pipe_build_write_buffer(wpipe, uio);
  964         PIPE_LOCK(wpipe);
  965         if (error) {
  966                 wpipe->pipe_state &= ~PIPE_DIRECTW;
  967                 pipeunlock(wpipe);
  968                 goto error1;
  969         }
  970 
  971         error = 0;
  972         while (!error && (wpipe->pipe_state & PIPE_DIRECTW)) {
  973                 if (wpipe->pipe_state & PIPE_EOF) {
  974                         pipe_destroy_write_buffer(wpipe);
  975                         pipeselwakeup(wpipe);
  976                         pipeunlock(wpipe);
  977                         error = EPIPE;
  978                         goto error1;
  979                 }
  980                 if (wpipe->pipe_state & PIPE_WANTR) {
  981                         wpipe->pipe_state &= ~PIPE_WANTR;
  982                         wakeup(wpipe);
  983                 }
  984                 pipeselwakeup(wpipe);
  985                 wpipe->pipe_state |= PIPE_WANTW;
  986                 pipeunlock(wpipe);
  987                 error = msleep(wpipe, PIPE_MTX(wpipe), PRIBIO | PCATCH,
  988                     "pipdwt", 0);
  989                 pipelock(wpipe, 0);
  990         }
  991 
  992         if (wpipe->pipe_state & PIPE_EOF)
  993                 error = EPIPE;
  994         if (wpipe->pipe_state & PIPE_DIRECTW) {
  995                 /*
  996                  * this bit of trickery substitutes a kernel buffer for
  997                  * the process that might be going away.
  998                  */
  999                 pipe_clone_write_buffer(wpipe);
 1000         } else {
 1001                 pipe_destroy_write_buffer(wpipe);
 1002         }
 1003         pipeunlock(wpipe);
 1004         return (error);
 1005 
 1006 error1:
 1007         wakeup(wpipe);
 1008         return (error);
 1009 }
 1010 #endif
 1011 
 1012 static int
 1013 pipe_write(struct file *fp, struct uio *uio, struct ucred *active_cred,
 1014     int flags, struct thread *td)
 1015 {
 1016         int error = 0;
 1017         int desiredsize;
 1018         ssize_t orig_resid;
 1019         struct pipe *wpipe, *rpipe;
 1020 
 1021         rpipe = fp->f_data;
 1022         wpipe = PIPE_PEER(rpipe);
 1023         PIPE_LOCK(rpipe);
 1024         error = pipelock(wpipe, 1);
 1025         if (error) {
 1026                 PIPE_UNLOCK(rpipe);
 1027                 return (error);
 1028         }
 1029         /*
 1030          * detect loss of pipe read side, issue SIGPIPE if lost.
 1031          */
 1032         if (wpipe->pipe_present != PIPE_ACTIVE ||
 1033             (wpipe->pipe_state & PIPE_EOF)) {
 1034                 pipeunlock(wpipe);
 1035                 PIPE_UNLOCK(rpipe);
 1036                 return (EPIPE);
 1037         }
 1038 #ifdef MAC
 1039         error = mac_pipe_check_write(active_cred, wpipe->pipe_pair);
 1040         if (error) {
 1041                 pipeunlock(wpipe);
 1042                 PIPE_UNLOCK(rpipe);
 1043                 return (error);
 1044         }
 1045 #endif
 1046         ++wpipe->pipe_busy;
 1047 
 1048         /* Choose a larger size if it's advantageous */
 1049         desiredsize = max(SMALL_PIPE_SIZE, wpipe->pipe_buffer.size);
 1050         while (desiredsize < wpipe->pipe_buffer.cnt + uio->uio_resid) {
 1051                 if (piperesizeallowed != 1)
 1052                         break;
 1053                 if (amountpipekva > maxpipekva / 2)
 1054                         break;
 1055                 if (desiredsize == BIG_PIPE_SIZE)
 1056                         break;
 1057                 desiredsize = desiredsize * 2;
 1058         }
 1059 
 1060         /* Choose a smaller size if we're in a OOM situation */
 1061         if ((amountpipekva > (3 * maxpipekva) / 4) &&
 1062                 (wpipe->pipe_buffer.size > SMALL_PIPE_SIZE) &&
 1063                 (wpipe->pipe_buffer.cnt <= SMALL_PIPE_SIZE) &&
 1064                 (piperesizeallowed == 1))
 1065                 desiredsize = SMALL_PIPE_SIZE;
 1066 
 1067         /* Resize if the above determined that a new size was necessary */
 1068         if ((desiredsize != wpipe->pipe_buffer.size) &&
 1069                 ((wpipe->pipe_state & PIPE_DIRECTW) == 0)) {
 1070                 PIPE_UNLOCK(wpipe);
 1071                 pipespace(wpipe, desiredsize);
 1072                 PIPE_LOCK(wpipe);
 1073         }
 1074         if (wpipe->pipe_buffer.size == 0) {
 1075                 /*
 1076                  * This can only happen for reverse direction use of pipes
 1077                  * in a complete OOM situation.
 1078                  */
 1079                 error = ENOMEM;
 1080                 --wpipe->pipe_busy;
 1081                 pipeunlock(wpipe);
 1082                 PIPE_UNLOCK(wpipe);
 1083                 return (error);
 1084         }
 1085 
 1086         pipeunlock(wpipe);
 1087 
 1088         orig_resid = uio->uio_resid;
 1089 
 1090         while (uio->uio_resid) {
 1091                 int space;
 1092 
 1093                 pipelock(wpipe, 0);
 1094                 if (wpipe->pipe_state & PIPE_EOF) {
 1095                         pipeunlock(wpipe);
 1096                         error = EPIPE;
 1097                         break;
 1098                 }
 1099 #ifndef PIPE_NODIRECT
 1100                 /*
 1101                  * If the transfer is large, we can gain performance if
 1102                  * we do process-to-process copies directly.
 1103                  * If the write is non-blocking, we don't use the
 1104                  * direct write mechanism.
 1105                  *
 1106                  * The direct write mechanism will detect the reader going
 1107                  * away on us.
 1108                  */
 1109                 if (uio->uio_segflg == UIO_USERSPACE &&
 1110                     uio->uio_iov->iov_len >= PIPE_MINDIRECT &&
 1111                     wpipe->pipe_buffer.size >= PIPE_MINDIRECT &&
 1112                     (fp->f_flag & FNONBLOCK) == 0) {
 1113                         pipeunlock(wpipe);
 1114                         error = pipe_direct_write(wpipe, uio);
 1115                         if (error)
 1116                                 break;
 1117                         continue;
 1118                 }
 1119 #endif
 1120 
 1121                 /*
 1122                  * Pipe buffered writes cannot be coincidental with
 1123                  * direct writes.  We wait until the currently executing
 1124                  * direct write is completed before we start filling the
 1125                  * pipe buffer.  We break out if a signal occurs or the
 1126                  * reader goes away.
 1127                  */
 1128                 if (wpipe->pipe_state & PIPE_DIRECTW) {
 1129                         if (wpipe->pipe_state & PIPE_WANTR) {
 1130                                 wpipe->pipe_state &= ~PIPE_WANTR;
 1131                                 wakeup(wpipe);
 1132                         }
 1133                         pipeselwakeup(wpipe);
 1134                         wpipe->pipe_state |= PIPE_WANTW;
 1135                         pipeunlock(wpipe);
 1136                         error = msleep(wpipe, PIPE_MTX(rpipe), PRIBIO | PCATCH,
 1137                             "pipbww", 0);
 1138                         if (error)
 1139                                 break;
 1140                         else
 1141                                 continue;
 1142                 }
 1143 
 1144                 space = wpipe->pipe_buffer.size - wpipe->pipe_buffer.cnt;
 1145 
 1146                 /* Writes of size <= PIPE_BUF must be atomic. */
 1147                 if ((space < uio->uio_resid) && (orig_resid <= PIPE_BUF))
 1148                         space = 0;
 1149 
 1150                 if (space > 0) {
 1151                         int size;       /* Transfer size */
 1152                         int segsize;    /* first segment to transfer */
 1153 
 1154                         /*
 1155                          * Transfer size is minimum of uio transfer
 1156                          * and free space in pipe buffer.
 1157                          */
 1158                         if (space > uio->uio_resid)
 1159                                 size = uio->uio_resid;
 1160                         else
 1161                                 size = space;
 1162                         /*
 1163                          * First segment to transfer is minimum of
 1164                          * transfer size and contiguous space in
 1165                          * pipe buffer.  If first segment to transfer
 1166                          * is less than the transfer size, we've got
 1167                          * a wraparound in the buffer.
 1168                          */
 1169                         segsize = wpipe->pipe_buffer.size -
 1170                                 wpipe->pipe_buffer.in;
 1171                         if (segsize > size)
 1172                                 segsize = size;
 1173 
 1174                         /* Transfer first segment */
 1175 
 1176                         PIPE_UNLOCK(rpipe);
 1177                         error = uiomove(&wpipe->pipe_buffer.buffer[wpipe->pipe_buffer.in],
 1178                                         segsize, uio);
 1179                         PIPE_LOCK(rpipe);
 1180 
 1181                         if (error == 0 && segsize < size) {
 1182                                 KASSERT(wpipe->pipe_buffer.in + segsize ==
 1183                                         wpipe->pipe_buffer.size,
 1184                                         ("Pipe buffer wraparound disappeared"));
 1185                                 /*
 1186                                  * Transfer remaining part now, to
 1187                                  * support atomic writes.  Wraparound
 1188                                  * happened.
 1189                                  */
 1190 
 1191                                 PIPE_UNLOCK(rpipe);
 1192                                 error = uiomove(
 1193                                     &wpipe->pipe_buffer.buffer[0],
 1194                                     size - segsize, uio);
 1195                                 PIPE_LOCK(rpipe);
 1196                         }
 1197                         if (error == 0) {
 1198                                 wpipe->pipe_buffer.in += size;
 1199                                 if (wpipe->pipe_buffer.in >=
 1200                                     wpipe->pipe_buffer.size) {
 1201                                         KASSERT(wpipe->pipe_buffer.in ==
 1202                                                 size - segsize +
 1203                                                 wpipe->pipe_buffer.size,
 1204                                                 ("Expected wraparound bad"));
 1205                                         wpipe->pipe_buffer.in = size - segsize;
 1206                                 }
 1207 
 1208                                 wpipe->pipe_buffer.cnt += size;
 1209                                 KASSERT(wpipe->pipe_buffer.cnt <=
 1210                                         wpipe->pipe_buffer.size,
 1211                                         ("Pipe buffer overflow"));
 1212                         }
 1213                         pipeunlock(wpipe);
 1214                         if (error != 0)
 1215                                 break;
 1216                 } else {
 1217                         /*
 1218                          * If the "read-side" has been blocked, wake it up now.
 1219                          */
 1220                         if (wpipe->pipe_state & PIPE_WANTR) {
 1221                                 wpipe->pipe_state &= ~PIPE_WANTR;
 1222                                 wakeup(wpipe);
 1223                         }
 1224 
 1225                         /*
 1226                          * don't block on non-blocking I/O
 1227                          */
 1228                         if (fp->f_flag & FNONBLOCK) {
 1229                                 error = EAGAIN;
 1230                                 pipeunlock(wpipe);
 1231                                 break;
 1232                         }
 1233 
 1234                         /*
 1235                          * We have no more space and have something to offer,
 1236                          * wake up select/poll.
 1237                          */
 1238                         pipeselwakeup(wpipe);
 1239 
 1240                         wpipe->pipe_state |= PIPE_WANTW;
 1241                         pipeunlock(wpipe);
 1242                         error = msleep(wpipe, PIPE_MTX(rpipe),
 1243                             PRIBIO | PCATCH, "pipewr", 0);
 1244                         if (error != 0)
 1245                                 break;
 1246                 }
 1247         }
 1248 
 1249         pipelock(wpipe, 0);
 1250         --wpipe->pipe_busy;
 1251 
 1252         if ((wpipe->pipe_busy == 0) && (wpipe->pipe_state & PIPE_WANT)) {
 1253                 wpipe->pipe_state &= ~(PIPE_WANT | PIPE_WANTR);
 1254                 wakeup(wpipe);
 1255         } else if (wpipe->pipe_buffer.cnt > 0) {
 1256                 /*
 1257                  * If we have put any characters in the buffer, we wake up
 1258                  * the reader.
 1259                  */
 1260                 if (wpipe->pipe_state & PIPE_WANTR) {
 1261                         wpipe->pipe_state &= ~PIPE_WANTR;
 1262                         wakeup(wpipe);
 1263                 }
 1264         }
 1265 
 1266         /*
 1267          * Don't return EPIPE if any byte was written.
 1268          * EINTR and other interrupts are handled by generic I/O layer.
 1269          * Do not pretend that I/O succeeded for obvious user error
 1270          * like EFAULT.
 1271          */
 1272         if (uio->uio_resid != orig_resid && error == EPIPE)
 1273                 error = 0;
 1274 
 1275         if (error == 0)
 1276                 vfs_timestamp(&wpipe->pipe_mtime);
 1277 
 1278         /*
 1279          * We have something to offer,
 1280          * wake up select/poll.
 1281          */
 1282         if (wpipe->pipe_buffer.cnt)
 1283                 pipeselwakeup(wpipe);
 1284 
 1285         pipeunlock(wpipe);
 1286         PIPE_UNLOCK(rpipe);
 1287         return (error);
 1288 }
 1289 
 1290 /* ARGSUSED */
 1291 static int
 1292 pipe_truncate(struct file *fp, off_t length, struct ucred *active_cred,
 1293     struct thread *td)
 1294 {
 1295         struct pipe *cpipe;
 1296         int error;
 1297 
 1298         cpipe = fp->f_data;
 1299         if (cpipe->pipe_state & PIPE_NAMED)
 1300                 error = vnops.fo_truncate(fp, length, active_cred, td);
 1301         else
 1302                 error = invfo_truncate(fp, length, active_cred, td);
 1303         return (error);
 1304 }
 1305 
 1306 /*
 1307  * we implement a very minimal set of ioctls for compatibility with sockets.
 1308  */
 1309 static int
 1310 pipe_ioctl(struct file *fp, u_long cmd, void *data, struct ucred *active_cred,
 1311     struct thread *td)
 1312 {
 1313         struct pipe *mpipe = fp->f_data;
 1314         int error;
 1315 
 1316         PIPE_LOCK(mpipe);
 1317 
 1318 #ifdef MAC
 1319         error = mac_pipe_check_ioctl(active_cred, mpipe->pipe_pair, cmd, data);
 1320         if (error) {
 1321                 PIPE_UNLOCK(mpipe);
 1322                 return (error);
 1323         }
 1324 #endif
 1325 
 1326         error = 0;
 1327         switch (cmd) {
 1328 
 1329         case FIONBIO:
 1330                 break;
 1331 
 1332         case FIOASYNC:
 1333                 if (*(int *)data) {
 1334                         mpipe->pipe_state |= PIPE_ASYNC;
 1335                 } else {
 1336                         mpipe->pipe_state &= ~PIPE_ASYNC;
 1337                 }
 1338                 break;
 1339 
 1340         case FIONREAD:
 1341                 if (!(fp->f_flag & FREAD)) {
 1342                         *(int *)data = 0;
 1343                         PIPE_UNLOCK(mpipe);
 1344                         return (0);
 1345                 }
 1346                 if (mpipe->pipe_state & PIPE_DIRECTW)
 1347                         *(int *)data = mpipe->pipe_map.cnt;
 1348                 else
 1349                         *(int *)data = mpipe->pipe_buffer.cnt;
 1350                 break;
 1351 
 1352         case FIOSETOWN:
 1353                 PIPE_UNLOCK(mpipe);
 1354                 error = fsetown(*(int *)data, &mpipe->pipe_sigio);
 1355                 goto out_unlocked;
 1356 
 1357         case FIOGETOWN:
 1358                 *(int *)data = fgetown(&mpipe->pipe_sigio);
 1359                 break;
 1360 
 1361         /* This is deprecated, FIOSETOWN should be used instead. */
 1362         case TIOCSPGRP:
 1363                 PIPE_UNLOCK(mpipe);
 1364                 error = fsetown(-(*(int *)data), &mpipe->pipe_sigio);
 1365                 goto out_unlocked;
 1366 
 1367         /* This is deprecated, FIOGETOWN should be used instead. */
 1368         case TIOCGPGRP:
 1369                 *(int *)data = -fgetown(&mpipe->pipe_sigio);
 1370                 break;
 1371 
 1372         default:
 1373                 error = ENOTTY;
 1374                 break;
 1375         }
 1376         PIPE_UNLOCK(mpipe);
 1377 out_unlocked:
 1378         return (error);
 1379 }
 1380 
 1381 static int
 1382 pipe_poll(struct file *fp, int events, struct ucred *active_cred,
 1383     struct thread *td)
 1384 {
 1385         struct pipe *rpipe;
 1386         struct pipe *wpipe;
 1387         int levents, revents;
 1388 #ifdef MAC
 1389         int error;
 1390 #endif
 1391 
 1392         revents = 0;
 1393         rpipe = fp->f_data;
 1394         wpipe = PIPE_PEER(rpipe);
 1395         PIPE_LOCK(rpipe);
 1396 #ifdef MAC
 1397         error = mac_pipe_check_poll(active_cred, rpipe->pipe_pair);
 1398         if (error)
 1399                 goto locked_error;
 1400 #endif
 1401         if (fp->f_flag & FREAD && events & (POLLIN | POLLRDNORM))
 1402                 if ((rpipe->pipe_state & PIPE_DIRECTW) ||
 1403                     (rpipe->pipe_buffer.cnt > 0))
 1404                         revents |= events & (POLLIN | POLLRDNORM);
 1405 
 1406         if (fp->f_flag & FWRITE && events & (POLLOUT | POLLWRNORM))
 1407                 if (wpipe->pipe_present != PIPE_ACTIVE ||
 1408                     (wpipe->pipe_state & PIPE_EOF) ||
 1409                     (((wpipe->pipe_state & PIPE_DIRECTW) == 0) &&
 1410                      ((wpipe->pipe_buffer.size - wpipe->pipe_buffer.cnt) >= PIPE_BUF ||
 1411                          wpipe->pipe_buffer.size == 0)))
 1412                         revents |= events & (POLLOUT | POLLWRNORM);
 1413 
 1414         levents = events &
 1415             (POLLIN | POLLINIGNEOF | POLLPRI | POLLRDNORM | POLLRDBAND);
 1416         if (rpipe->pipe_state & PIPE_NAMED && fp->f_flag & FREAD && levents &&
 1417             fp->f_seqcount == rpipe->pipe_wgen)
 1418                 events |= POLLINIGNEOF;
 1419 
 1420         if ((events & POLLINIGNEOF) == 0) {
 1421                 if (rpipe->pipe_state & PIPE_EOF) {
 1422                         revents |= (events & (POLLIN | POLLRDNORM));
 1423                         if (wpipe->pipe_present != PIPE_ACTIVE ||
 1424                             (wpipe->pipe_state & PIPE_EOF))
 1425                                 revents |= POLLHUP;
 1426                 }
 1427         }
 1428 
 1429         if (revents == 0) {
 1430                 if (fp->f_flag & FREAD && events & (POLLIN | POLLRDNORM)) {
 1431                         selrecord(td, &rpipe->pipe_sel);
 1432                         if (SEL_WAITING(&rpipe->pipe_sel))
 1433                                 rpipe->pipe_state |= PIPE_SEL;
 1434                 }
 1435 
 1436                 if (fp->f_flag & FWRITE && events & (POLLOUT | POLLWRNORM)) {
 1437                         selrecord(td, &wpipe->pipe_sel);
 1438                         if (SEL_WAITING(&wpipe->pipe_sel))
 1439                                 wpipe->pipe_state |= PIPE_SEL;
 1440                 }
 1441         }
 1442 #ifdef MAC
 1443 locked_error:
 1444 #endif
 1445         PIPE_UNLOCK(rpipe);
 1446 
 1447         return (revents);
 1448 }
 1449 
 1450 /*
 1451  * We shouldn't need locks here as we're doing a read and this should
 1452  * be a natural race.
 1453  */
 1454 static int
 1455 pipe_stat(struct file *fp, struct stat *ub, struct ucred *active_cred,
 1456     struct thread *td)
 1457 {
 1458         struct pipe *pipe;
 1459 #ifdef MAC
 1460         int error;
 1461 #endif
 1462 
 1463         pipe = fp->f_data;
 1464         PIPE_LOCK(pipe);
 1465 #ifdef MAC
 1466         error = mac_pipe_check_stat(active_cred, pipe->pipe_pair);
 1467         if (error) {
 1468                 PIPE_UNLOCK(pipe);
 1469                 return (error);
 1470         }
 1471 #endif
 1472 
 1473         /* For named pipes ask the underlying filesystem. */
 1474         if (pipe->pipe_state & PIPE_NAMED) {
 1475                 PIPE_UNLOCK(pipe);
 1476                 return (vnops.fo_stat(fp, ub, active_cred, td));
 1477         }
 1478 
 1479         PIPE_UNLOCK(pipe);
 1480 
 1481         bzero(ub, sizeof(*ub));
 1482         ub->st_mode = S_IFIFO;
 1483         ub->st_blksize = PAGE_SIZE;
 1484         if (pipe->pipe_state & PIPE_DIRECTW)
 1485                 ub->st_size = pipe->pipe_map.cnt;
 1486         else
 1487                 ub->st_size = pipe->pipe_buffer.cnt;
 1488         ub->st_blocks = howmany(ub->st_size, ub->st_blksize);
 1489         ub->st_atim = pipe->pipe_atime;
 1490         ub->st_mtim = pipe->pipe_mtime;
 1491         ub->st_ctim = pipe->pipe_ctime;
 1492         ub->st_uid = fp->f_cred->cr_uid;
 1493         ub->st_gid = fp->f_cred->cr_gid;
 1494         ub->st_dev = pipedev_ino;
 1495         ub->st_ino = pipe->pipe_ino;
 1496         /*
 1497          * Left as 0: st_nlink, st_rdev, st_flags, st_gen.
 1498          */
 1499         return (0);
 1500 }
 1501 
 1502 /* ARGSUSED */
 1503 static int
 1504 pipe_close(struct file *fp, struct thread *td)
 1505 {
 1506 
 1507         if (fp->f_vnode != NULL) 
 1508                 return vnops.fo_close(fp, td);
 1509         fp->f_ops = &badfileops;
 1510         pipe_dtor(fp->f_data);
 1511         fp->f_data = NULL;
 1512         return (0);
 1513 }
 1514 
 1515 static int
 1516 pipe_chmod(struct file *fp, mode_t mode, struct ucred *active_cred, struct thread *td)
 1517 {
 1518         struct pipe *cpipe;
 1519         int error;
 1520 
 1521         cpipe = fp->f_data;
 1522         if (cpipe->pipe_state & PIPE_NAMED)
 1523                 error = vn_chmod(fp, mode, active_cred, td);
 1524         else
 1525                 error = invfo_chmod(fp, mode, active_cred, td);
 1526         return (error);
 1527 }
 1528 
 1529 static int
 1530 pipe_chown(struct file *fp, uid_t uid, gid_t gid, struct ucred *active_cred,
 1531     struct thread *td)
 1532 {
 1533         struct pipe *cpipe;
 1534         int error;
 1535 
 1536         cpipe = fp->f_data;
 1537         if (cpipe->pipe_state & PIPE_NAMED)
 1538                 error = vn_chown(fp, uid, gid, active_cred, td);
 1539         else
 1540                 error = invfo_chown(fp, uid, gid, active_cred, td);
 1541         return (error);
 1542 }
 1543 
 1544 static int
 1545 pipe_fill_kinfo(struct file *fp, struct kinfo_file *kif, struct filedesc *fdp)
 1546 {
 1547         struct pipe *pi;
 1548 
 1549         if (fp->f_type == DTYPE_FIFO)
 1550                 return (vn_fill_kinfo(fp, kif, fdp));
 1551         kif->kf_type = KF_TYPE_PIPE;
 1552         pi = fp->f_data;
 1553         kif->kf_un.kf_pipe.kf_pipe_addr = (uintptr_t)pi;
 1554         kif->kf_un.kf_pipe.kf_pipe_peer = (uintptr_t)pi->pipe_peer;
 1555         kif->kf_un.kf_pipe.kf_pipe_buffer_cnt = pi->pipe_buffer.cnt;
 1556         return (0);
 1557 }
 1558 
 1559 static void
 1560 pipe_free_kmem(struct pipe *cpipe)
 1561 {
 1562 
 1563         KASSERT(!mtx_owned(PIPE_MTX(cpipe)),
 1564             ("pipe_free_kmem: pipe mutex locked"));
 1565 
 1566         if (cpipe->pipe_buffer.buffer != NULL) {
 1567                 atomic_subtract_long(&amountpipekva, cpipe->pipe_buffer.size);
 1568                 vm_map_remove(pipe_map,
 1569                     (vm_offset_t)cpipe->pipe_buffer.buffer,
 1570                     (vm_offset_t)cpipe->pipe_buffer.buffer + cpipe->pipe_buffer.size);
 1571                 cpipe->pipe_buffer.buffer = NULL;
 1572         }
 1573 #ifndef PIPE_NODIRECT
 1574         {
 1575                 cpipe->pipe_map.cnt = 0;
 1576                 cpipe->pipe_map.pos = 0;
 1577                 cpipe->pipe_map.npages = 0;
 1578         }
 1579 #endif
 1580 }
 1581 
 1582 /*
 1583  * shutdown the pipe
 1584  */
 1585 static void
 1586 pipeclose(struct pipe *cpipe)
 1587 {
 1588         struct pipepair *pp;
 1589         struct pipe *ppipe;
 1590 
 1591         KASSERT(cpipe != NULL, ("pipeclose: cpipe == NULL"));
 1592 
 1593         PIPE_LOCK(cpipe);
 1594         pipelock(cpipe, 0);
 1595         pp = cpipe->pipe_pair;
 1596 
 1597         pipeselwakeup(cpipe);
 1598 
 1599         /*
 1600          * If the other side is blocked, wake it up saying that
 1601          * we want to close it down.
 1602          */
 1603         cpipe->pipe_state |= PIPE_EOF;
 1604         while (cpipe->pipe_busy) {
 1605                 wakeup(cpipe);
 1606                 cpipe->pipe_state |= PIPE_WANT;
 1607                 pipeunlock(cpipe);
 1608                 msleep(cpipe, PIPE_MTX(cpipe), PRIBIO, "pipecl", 0);
 1609                 pipelock(cpipe, 0);
 1610         }
 1611 
 1612 
 1613         /*
 1614          * Disconnect from peer, if any.
 1615          */
 1616         ppipe = cpipe->pipe_peer;
 1617         if (ppipe->pipe_present == PIPE_ACTIVE) {
 1618                 pipeselwakeup(ppipe);
 1619 
 1620                 ppipe->pipe_state |= PIPE_EOF;
 1621                 wakeup(ppipe);
 1622                 KNOTE_LOCKED(&ppipe->pipe_sel.si_note, 0);
 1623         }
 1624 
 1625         /*
 1626          * Mark this endpoint as free.  Release kmem resources.  We
 1627          * don't mark this endpoint as unused until we've finished
 1628          * doing that, or the pipe might disappear out from under
 1629          * us.
 1630          */
 1631         PIPE_UNLOCK(cpipe);
 1632         pipe_free_kmem(cpipe);
 1633         PIPE_LOCK(cpipe);
 1634         cpipe->pipe_present = PIPE_CLOSING;
 1635         pipeunlock(cpipe);
 1636 
 1637         /*
 1638          * knlist_clear() may sleep dropping the PIPE_MTX. Set the
 1639          * PIPE_FINALIZED, that allows other end to free the
 1640          * pipe_pair, only after the knotes are completely dismantled.
 1641          */
 1642         knlist_clear(&cpipe->pipe_sel.si_note, 1);
 1643         cpipe->pipe_present = PIPE_FINALIZED;
 1644         seldrain(&cpipe->pipe_sel);
 1645         knlist_destroy(&cpipe->pipe_sel.si_note);
 1646 
 1647         /*
 1648          * If both endpoints are now closed, release the memory for the
 1649          * pipe pair.  If not, unlock.
 1650          */
 1651         if (ppipe->pipe_present == PIPE_FINALIZED) {
 1652                 PIPE_UNLOCK(cpipe);
 1653 #ifdef MAC
 1654                 mac_pipe_destroy(pp);
 1655 #endif
 1656                 uma_zfree(pipe_zone, cpipe->pipe_pair);
 1657         } else
 1658                 PIPE_UNLOCK(cpipe);
 1659 }
 1660 
 1661 /*ARGSUSED*/
 1662 static int
 1663 pipe_kqfilter(struct file *fp, struct knote *kn)
 1664 {
 1665         struct pipe *cpipe;
 1666 
 1667         /*
 1668          * If a filter is requested that is not supported by this file
 1669          * descriptor, don't return an error, but also don't ever generate an
 1670          * event.
 1671          */
 1672         if ((kn->kn_filter == EVFILT_READ) && !(fp->f_flag & FREAD)) {
 1673                 kn->kn_fop = &pipe_nfiltops;
 1674                 return (0);
 1675         }
 1676         if ((kn->kn_filter == EVFILT_WRITE) && !(fp->f_flag & FWRITE)) {
 1677                 kn->kn_fop = &pipe_nfiltops;
 1678                 return (0);
 1679         }
 1680         cpipe = fp->f_data;
 1681         PIPE_LOCK(cpipe);
 1682         switch (kn->kn_filter) {
 1683         case EVFILT_READ:
 1684                 kn->kn_fop = &pipe_rfiltops;
 1685                 break;
 1686         case EVFILT_WRITE:
 1687                 kn->kn_fop = &pipe_wfiltops;
 1688                 if (cpipe->pipe_peer->pipe_present != PIPE_ACTIVE) {
 1689                         /* other end of pipe has been closed */
 1690                         PIPE_UNLOCK(cpipe);
 1691                         return (EPIPE);
 1692                 }
 1693                 cpipe = PIPE_PEER(cpipe);
 1694                 break;
 1695         default:
 1696                 PIPE_UNLOCK(cpipe);
 1697                 return (EINVAL);
 1698         }
 1699 
 1700         kn->kn_hook = cpipe; 
 1701         knlist_add(&cpipe->pipe_sel.si_note, kn, 1);
 1702         PIPE_UNLOCK(cpipe);
 1703         return (0);
 1704 }
 1705 
 1706 static void
 1707 filt_pipedetach(struct knote *kn)
 1708 {
 1709         struct pipe *cpipe = kn->kn_hook;
 1710 
 1711         PIPE_LOCK(cpipe);
 1712         knlist_remove(&cpipe->pipe_sel.si_note, kn, 1);
 1713         PIPE_UNLOCK(cpipe);
 1714 }
 1715 
 1716 /*ARGSUSED*/
 1717 static int
 1718 filt_piperead(struct knote *kn, long hint)
 1719 {
 1720         struct pipe *rpipe = kn->kn_hook;
 1721         struct pipe *wpipe = rpipe->pipe_peer;
 1722         int ret;
 1723 
 1724         PIPE_LOCK_ASSERT(rpipe, MA_OWNED);
 1725         kn->kn_data = rpipe->pipe_buffer.cnt;
 1726         if ((kn->kn_data == 0) && (rpipe->pipe_state & PIPE_DIRECTW))
 1727                 kn->kn_data = rpipe->pipe_map.cnt;
 1728 
 1729         if ((rpipe->pipe_state & PIPE_EOF) ||
 1730             wpipe->pipe_present != PIPE_ACTIVE ||
 1731             (wpipe->pipe_state & PIPE_EOF)) {
 1732                 kn->kn_flags |= EV_EOF;
 1733                 return (1);
 1734         }
 1735         ret = kn->kn_data > 0;
 1736         return ret;
 1737 }
 1738 
 1739 /*ARGSUSED*/
 1740 static int
 1741 filt_pipewrite(struct knote *kn, long hint)
 1742 {
 1743         struct pipe *wpipe;
 1744 
 1745         /*
 1746          * If this end of the pipe is closed, the knote was removed from the
 1747          * knlist and the list lock (i.e., the pipe lock) is therefore not held.
 1748          */
 1749         wpipe = kn->kn_hook;
 1750         if (wpipe->pipe_present != PIPE_ACTIVE ||
 1751             (wpipe->pipe_state & PIPE_EOF)) {
 1752                 kn->kn_data = 0;
 1753                 kn->kn_flags |= EV_EOF;
 1754                 return (1);
 1755         }
 1756         PIPE_LOCK_ASSERT(wpipe, MA_OWNED);
 1757         kn->kn_data = (wpipe->pipe_buffer.size > 0) ?
 1758             (wpipe->pipe_buffer.size - wpipe->pipe_buffer.cnt) : PIPE_BUF;
 1759         if (wpipe->pipe_state & PIPE_DIRECTW)
 1760                 kn->kn_data = 0;
 1761 
 1762         return (kn->kn_data >= PIPE_BUF);
 1763 }
 1764 
 1765 static void
 1766 filt_pipedetach_notsup(struct knote *kn)
 1767 {
 1768 
 1769 }
 1770 
 1771 static int
 1772 filt_pipenotsup(struct knote *kn, long hint)
 1773 {
 1774 
 1775         return (0);
 1776 }

Cache object: 1e6741f9903bfe1e87f5f1d4ffc2cfce


[ 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.