sys/kern/sys_pipe.c

   1 /*
   2  * Copyright (c) 1996 John S. Dyson
   3  * All rights reserved.
   4  *
   5  * Redistribution and use in source and binary forms, with or without
   6  * modification, are permitted provided that the following conditions
   7  * are met:
   8  * 1. Redistributions of source code must retain the above copyright
   9  *    notice immediately at the beginning of the file, without modification,
  10  *    this list of conditions, and the following disclaimer.
  11  * 2. Redistributions in binary form must reproduce the above copyright
  12  *    notice, this list of conditions and the following disclaimer in the
  13  *    documentation and/or other materials provided with the distribution.
  14  * 3. Absolutely no warranty of function or purpose is made by the author
  15  *    John S. Dyson.
  16  * 4. Modifications may be freely made to this file if the above conditions
  17  *    are met.
  18  *
  19  * $FreeBSD: src/sys/kern/sys_pipe.c,v 1.60.2.13 2002/08/05 15:05:15 des Exp $
  20  * $DragonFly: src/sys/kern/sys_pipe.c,v 1.14 2004/02/20 17:11:07 dillon Exp $
  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, it is fully mapped and wired into the kernel, and
  36  * the receiving process can copy it directly from the pages in the sending
  37  * 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.  PIPE_SIZE is constrained by the
  50  * amount of kernel virtual memory.
  51  */
  52
  53 #include <sys/param.h>
  54 #include <sys/systm.h>
  55 #include <sys/kernel.h>
  56 #include <sys/proc.h>
  57 #include <sys/fcntl.h>
  58 #include <sys/file.h>
  59 #include <sys/filedesc.h>
  60 #include <sys/filio.h>
  61 #include <sys/ttycom.h>
  62 #include <sys/stat.h>
  63 #include <sys/poll.h>
  64 #include <sys/select.h>
  65 #include <sys/signalvar.h>
  66 #include <sys/sysproto.h>
  67 #include <sys/pipe.h>
  68 #include <sys/vnode.h>
  69 #include <sys/uio.h>
  70 #include <sys/event.h>
  71 #include <sys/globaldata.h>
  72 #include <sys/module.h>
  73 #include <sys/malloc.h>
  74 #include <sys/sysctl.h>
  75
  76 #include <vm/vm.h>
  77 #include <vm/vm_param.h>
  78 #include <sys/lock.h>
  79 #include <vm/vm_object.h>
  80 #include <vm/vm_kern.h>
  81 #include <vm/vm_extern.h>
  82 #include <vm/pmap.h>
  83 #include <vm/vm_map.h>
  84 #include <vm/vm_page.h>
  85 #include <vm/vm_zone.h>
  86
  87 #include <sys/file2.h>
  88
  89 /*
  90  * Use this define if you want to disable *fancy* VM things.  Expect an
  91  * approx 30% decrease in transfer rate.  This could be useful for
  92  * NetBSD or OpenBSD.
  93  */
  94 /* #define PIPE_NODIRECT */
  95
  96 /*
  97  * interfaces to the outside world
  98  */
  99 static int pipe_read (struct file *fp, struct uio *uio,
 100                 struct ucred *cred, int flags, struct thread *td);
 101 static int pipe_write (struct file *fp, struct uio *uio,
 102                 struct ucred *cred, int flags, struct thread *td);
 103 static int pipe_close (struct file *fp, struct thread *td);
 104 static int pipe_poll (struct file *fp, int events, struct ucred *cred,
 105                 struct thread *td);
 106 static int pipe_kqfilter (struct file *fp, struct knote *kn);
 107 static int pipe_stat (struct file *fp, struct stat *sb, struct thread *td);
 108 static int pipe_ioctl (struct file *fp, u_long cmd, caddr_t data, struct thread *td);
 109
 110 static struct fileops pipeops = {
 111         NULL,   /* port */
 112         0,      /* autoq */
 113         pipe_read, pipe_write, pipe_ioctl, pipe_poll, pipe_kqfilter,
 114         pipe_stat, pipe_close
 115 };
 116
 117 static void     filt_pipedetach(struct knote *kn);
 118 static int      filt_piperead(struct knote *kn, long hint);
 119 static int      filt_pipewrite(struct knote *kn, long hint);
 120
 121 static struct filterops pipe_rfiltops =
 122         { 1, NULL, filt_pipedetach, filt_piperead };
 123 static struct filterops pipe_wfiltops =
 124         { 1, NULL, filt_pipedetach, filt_pipewrite };
 125
 126 MALLOC_DEFINE(M_PIPE, "pipe", "pipe structures");
 127
 128 /*
 129  * Default pipe buffer size(s), this can be kind-of large now because pipe
 130  * space is pageable.  The pipe code will try to maintain locality of
 131  * reference for performance reasons, so small amounts of outstanding I/O
 132  * will not wipe the cache.
 133  */
 134 #define MINPIPESIZE (PIPE_SIZE/3)
 135 #define MAXPIPESIZE (2*PIPE_SIZE/3)
 136
 137 /*
 138  * Maximum amount of kva for pipes -- this is kind-of a soft limit, but
 139  * is there so that on large systems, we don't exhaust it.
 140  */
 141 #define MAXPIPEKVA (8*1024*1024)
 142
 143 /*
 144  * Limit for direct transfers, we cannot, of course limit
 145  * the amount of kva for pipes in general though.
 146  */
 147 #define LIMITPIPEKVA (16*1024*1024)
 148
 149 /*
 150  * Limit the number of "big" pipes
 151  */
 152 #define LIMITBIGPIPES   32
 153 #define PIPEQ_MAX_CACHE 16      /* per-cpu pipe structure cache */
 154
 155 static int pipe_maxbig = LIMITBIGPIPES;
 156 static int pipe_maxcache = PIPEQ_MAX_CACHE;
 157 static int pipe_nbig;
 158 static int pipe_kva;
 159 static int pipe_bcache_alloc;
 160 static int pipe_bkmem_alloc;
 161 static int pipe_dcache_alloc;
 162 static int pipe_dkmem_alloc;
 163
 164 SYSCTL_NODE(_kern, OID_AUTO, pipe, CTLFLAG_RW, 0, "Pipe operation");
 165 SYSCTL_INT(_kern_pipe, OID_AUTO, nbig,
 166         CTLFLAG_RD, &pipe_nbig, 0, "numer of big pipes allocated");
 167 SYSCTL_INT(_kern_pipe, OID_AUTO, kva,
 168         CTLFLAG_RD, &pipe_kva, 0, "kva reserved by pipes");
 169 SYSCTL_INT(_kern_pipe, OID_AUTO, maxcache,
 170         CTLFLAG_RW, &pipe_maxcache, 0, "max pipes cached per-cpu");
 171 SYSCTL_INT(_kern_pipe, OID_AUTO, maxbig,
 172         CTLFLAG_RW, &pipe_maxbig, 0, "max number of big pipes");
 173 #if !defined(NO_PIPE_SYSCTL_STATS)
 174 SYSCTL_INT(_kern_pipe, OID_AUTO, bcache_alloc,
 175         CTLFLAG_RW, &pipe_bcache_alloc, 0, "pipe buffer from pcpu cache");
 176 SYSCTL_INT(_kern_pipe, OID_AUTO, dcache_alloc,
 177         CTLFLAG_RW, &pipe_dcache_alloc, 0, "pipe direct buf from pcpu cache");
 178 SYSCTL_INT(_kern_pipe, OID_AUTO, bkmem_alloc,
 179         CTLFLAG_RW, &pipe_bkmem_alloc, 0, "pipe buffer from kmem");
 180 SYSCTL_INT(_kern_pipe, OID_AUTO, dkmem_alloc,
 181         CTLFLAG_RW, &pipe_dkmem_alloc, 0, "pipe direct buf from kmem");
 182 #endif
 183
 184 static void pipeclose (struct pipe *cpipe);
 185 static void pipe_free_kmem (struct pipe *cpipe);
 186 static int pipe_create (struct pipe **cpipep);
 187 static __inline int pipelock (struct pipe *cpipe, int catch);
 188 static __inline void pipeunlock (struct pipe *cpipe);
 189 static __inline void pipeselwakeup (struct pipe *cpipe);
 190 #ifndef PIPE_NODIRECT
 191 static int pipe_build_write_buffer (struct pipe *wpipe, struct uio *uio);
 192 static void pipe_destroy_write_buffer (struct pipe *wpipe);
 193 static int pipe_direct_write (struct pipe *wpipe, struct uio *uio);
 194 static void pipe_clone_write_buffer (struct pipe *wpipe);
 195 #endif
 196 static int pipespace (struct pipe *cpipe, int size);
 197
 198 /*
 199  * The pipe system call for the DTYPE_PIPE type of pipes
 200  *
 201  * pipe_ARgs(int dummy)
 202  */
 203
 204 /* ARGSUSED */
 205 int
 206 pipe(struct pipe_args *uap)
 207 {
 208         struct thread *td = curthread;
 209         struct proc *p = td->td_proc;
 210         struct filedesc *fdp;
 211         struct file *rf, *wf;
 212         struct pipe *rpipe, *wpipe;
 213         int fd1, fd2, error;
 214
 215         KKASSERT(p);
 216         fdp = p->p_fd;
 217
 218         rpipe = wpipe = NULL;
 219         if (pipe_create(&rpipe) || pipe_create(&wpipe)) {
 220                 pipeclose(rpipe);
 221                 pipeclose(wpipe);
 222                 return (ENFILE);
 223         }
 224
 225         rpipe->pipe_state |= PIPE_DIRECTOK;
 226         wpipe->pipe_state |= PIPE_DIRECTOK;
 227
 228         error = falloc(p, &rf, &fd1);
 229         if (error) {
 230                 pipeclose(rpipe);
 231                 pipeclose(wpipe);
 232                 return (error);
 233         }
 234         fhold(rf);
 235         uap->sysmsg_fds[0] = fd1;
 236
 237         /*
 238          * Warning: once we've gotten past allocation of the fd for the
 239          * read-side, we can only drop the read side via fdrop() in order
 240          * to avoid races against processes which manage to dup() the read
 241          * side while we are blocked trying to allocate the write side.
 242          */
 243         rf->f_flag = FREAD | FWRITE;
 244         rf->f_type = DTYPE_PIPE;
 245         rf->f_data = (caddr_t)rpipe;
 246         rf->f_ops = &pipeops;
 247         error = falloc(p, &wf, &fd2);
 248         if (error) {
 249                 if (fdp->fd_ofiles[fd1] == rf) {
 250                         fdp->fd_ofiles[fd1] = NULL;
 251                         fdrop(rf, td);
 252                 }
 253                 fdrop(rf, td);
 254                 /* rpipe has been closed by fdrop(). */
 255                 pipeclose(wpipe);
 256                 return (error);
 257         }
 258         wf->f_flag = FREAD | FWRITE;
 259         wf->f_type = DTYPE_PIPE;
 260         wf->f_data = (caddr_t)wpipe;
 261         wf->f_ops = &pipeops;
 262         uap->sysmsg_fds[1] = fd2;
 263
 264         rpipe->pipe_peer = wpipe;
 265         wpipe->pipe_peer = rpipe;
 266         fdrop(rf, td);
 267
 268         return (0);
 269 }
 270
 271 /*
 272  * Allocate kva for pipe circular buffer, the space is pageable
 273  * This routine will 'realloc' the size of a pipe safely, if it fails
 274  * it will retain the old buffer.
 275  * If it fails it will return ENOMEM.
 276  */
 277 static int
 278 pipespace(struct pipe *cpipe, int size)
 279 {
 280         struct vm_object *object;
 281         caddr_t buffer;
 282         int npages, error;
 283
 284         npages = round_page(size) / PAGE_SIZE;
 285         object = cpipe->pipe_buffer.object;
 286
 287         /*
 288          * [re]create the object if necessary and reserve space for it
 289          * in the kernel_map.  The object and memory are pageable.  On
 290          * success, free the old resources before assigning the new
 291          * ones.
 292          */
 293         if (object == NULL || object->size != npages) {
 294                 object = vm_object_allocate(OBJT_DEFAULT, npages);
 295                 buffer = (caddr_t) vm_map_min(kernel_map);
 296
 297                 error = vm_map_find(kernel_map, object, 0,
 298                         (vm_offset_t *) &buffer, size, 1,
 299                         VM_PROT_ALL, VM_PROT_ALL, 0);
 300
 301                 if (error != KERN_SUCCESS) {
 302                         vm_object_deallocate(object);
 303                         return (ENOMEM);
 304                 }
 305                 pipe_kva += size;
 306                 pipe_free_kmem(cpipe);
 307                 cpipe->pipe_buffer.object = object;
 308                 cpipe->pipe_buffer.buffer = buffer;
 309                 cpipe->pipe_buffer.size = size;
 310                 ++pipe_bkmem_alloc;
 311         } else {
 312                 ++pipe_bcache_alloc;
 313                 if (cpipe->pipe_map.kva)
 314                         ++pipe_dcache_alloc;
 315         }
 316         cpipe->pipe_buffer.in = 0;
 317         cpipe->pipe_buffer.out = 0;
 318         cpipe->pipe_buffer.cnt = 0;
 319         return (0);
 320 }
 321
 322 /*
 323  * Initialize and allocate VM and memory for pipe, pulling the pipe from
 324  * our per-cpu cache if possible.  For now make sure it is sized for the
 325  * smaller PIPE_SIZE default.
 326  */
 327 static int
 328 pipe_create(cpipep)
 329         struct pipe **cpipep;
 330 {
 331         globaldata_t gd = mycpu;
 332         struct pipe *cpipe;
 333         int error;
 334
 335         if ((cpipe = gd->gd_pipeq) != NULL) {
 336                 gd->gd_pipeq = cpipe->pipe_peer;
 337                 --gd->gd_pipeqcount;
 338                 cpipe->pipe_peer = NULL;
 339         } else {
 340                 cpipe = malloc(sizeof(struct pipe), M_PIPE, M_WAITOK|M_ZERO);
 341         }
 342         *cpipep = cpipe;
 343         if ((error = pipespace(cpipe, PIPE_SIZE)) != 0)
 344                 return (error);
 345         vfs_timestamp(&cpipe->pipe_ctime);
 346         cpipe->pipe_atime = cpipe->pipe_ctime;
 347         cpipe->pipe_mtime = cpipe->pipe_ctime;
 348         return (0);
 349 }
 350
 351
 352 /*
 353  * lock a pipe for I/O, blocking other access
 354  */
 355 static __inline int
 356 pipelock(cpipe, catch)
 357         struct pipe *cpipe;
 358         int catch;
 359 {
 360         int error;
 361
 362         while (cpipe->pipe_state & PIPE_LOCK) {
 363                 cpipe->pipe_state |= PIPE_LWANT;
 364                 error = tsleep(cpipe, (catch ? PCATCH : 0), "pipelk", 0);
 365                 if (error != 0)
 366                         return (error);
 367         }
 368         cpipe->pipe_state |= PIPE_LOCK;
 369         return (0);
 370 }
 371
 372 /*
 373  * unlock a pipe I/O lock
 374  */
 375 static __inline void
 376 pipeunlock(cpipe)
 377         struct pipe *cpipe;
 378 {
 379
 380         cpipe->pipe_state &= ~PIPE_LOCK;
 381         if (cpipe->pipe_state & PIPE_LWANT) {
 382                 cpipe->pipe_state &= ~PIPE_LWANT;
 383                 wakeup(cpipe);
 384         }
 385 }
 386
 387 static __inline void
 388 pipeselwakeup(cpipe)
 389         struct pipe *cpipe;
 390 {
 391
 392         if (cpipe->pipe_state & PIPE_SEL) {
 393                 cpipe->pipe_state &= ~PIPE_SEL;
 394                 selwakeup(&cpipe->pipe_sel);
 395         }
 396         if ((cpipe->pipe_state & PIPE_ASYNC) && cpipe->pipe_sigio)
 397                 pgsigio(cpipe->pipe_sigio, SIGIO, 0);
 398         KNOTE(&cpipe->pipe_sel.si_note, 0);
 399 }
 400
 401 /* ARGSUSED */
 402 static int
 403 pipe_read(struct file *fp, struct uio *uio, struct ucred *cred,
 404         int flags, struct thread *td)
 405 {
 406         struct pipe *rpipe = (struct pipe *) fp->f_data;
 407         int error;
 408         int nread = 0;
 409         u_int size;
 410
 411         ++rpipe->pipe_busy;
 412         error = pipelock(rpipe, 1);
 413         if (error)
 414                 goto unlocked_error;
 415
 416         while (uio->uio_resid) {
 417                 /*
 418                  * normal pipe buffer receive
 419                  */
 420                 if (rpipe->pipe_buffer.cnt > 0) {
 421                         size = rpipe->pipe_buffer.size - rpipe->pipe_buffer.out;
 422                         if (size > rpipe->pipe_buffer.cnt)
 423                                 size = rpipe->pipe_buffer.cnt;
 424                         if (size > (u_int) uio->uio_resid)
 425                                 size = (u_int) uio->uio_resid;
 426
 427                         error = uiomove(&rpipe->pipe_buffer.buffer[rpipe->pipe_buffer.out],
 428                                         size, uio);
 429                         if (error)
 430                                 break;
 431
 432                         rpipe->pipe_buffer.out += size;
 433                         if (rpipe->pipe_buffer.out >= rpipe->pipe_buffer.size)
 434                                 rpipe->pipe_buffer.out = 0;
 435
 436                         rpipe->pipe_buffer.cnt -= size;
 437
 438                         /*
 439                          * If there is no more to read in the pipe, reset
 440                          * its pointers to the beginning.  This improves
 441                          * cache hit stats.
 442                          */
 443                         if (rpipe->pipe_buffer.cnt == 0) {
 444                                 rpipe->pipe_buffer.in = 0;
 445                                 rpipe->pipe_buffer.out = 0;
 446                         }
 447                         nread += size;
 448 #ifndef PIPE_NODIRECT
 449                 /*
 450                  * Direct copy, bypassing a kernel buffer.
 451                  */
 452                 } else if ((size = rpipe->pipe_map.cnt) &&
 453                            (rpipe->pipe_state & PIPE_DIRECTW)) {
 454                         caddr_t va;
 455                         if (size > (u_int) uio->uio_resid)
 456                                 size = (u_int) uio->uio_resid;
 457
 458                         va = (caddr_t) rpipe->pipe_map.kva +
 459                             rpipe->pipe_map.pos;
 460                         error = uiomove(va, size, uio);
 461                         if (error)
 462                                 break;
 463                         nread += size;
 464                         rpipe->pipe_map.pos += size;
 465                         rpipe->pipe_map.cnt -= size;
 466                         if (rpipe->pipe_map.cnt == 0) {
 467                                 rpipe->pipe_state &= ~PIPE_DIRECTW;
 468                                 wakeup(rpipe);
 469                         }
 470 #endif
 471                 } else {
 472                         /*
 473                          * detect EOF condition
 474                          * read returns 0 on EOF, no need to set error
 475                          */
 476                         if (rpipe->pipe_state & PIPE_EOF)
 477                                 break;
 478
 479                         /*
 480                          * If the "write-side" has been blocked, wake it up now.
 481                          */
 482                         if (rpipe->pipe_state & PIPE_WANTW) {
 483                                 rpipe->pipe_state &= ~PIPE_WANTW;
 484                                 wakeup(rpipe);
 485                         }
 486
 487                         /*
 488                          * Break if some data was read.
 489                          */
 490                         if (nread > 0)
 491                                 break;
 492
 493                         /*
 494                          * Unlock the pipe buffer for our remaining processing.  We
 495                          * will either break out with an error or we will sleep and
 496                          * relock to loop.
 497                          */
 498                         pipeunlock(rpipe);
 499
 500                         /*
 501                          * Handle non-blocking mode operation or
 502                          * wait for more data.
 503                          */
 504                         if (fp->f_flag & FNONBLOCK) {
 505                                 error = EAGAIN;
 506                         } else {
 507                                 rpipe->pipe_state |= PIPE_WANTR;
 508                                 if ((error = tsleep(rpipe, PCATCH,
 509                                     "piperd", 0)) == 0) {
 510                                         error = pipelock(rpipe, 1);
 511                                 }
 512                         }
 513                         if (error)
 514                                 goto unlocked_error;
 515                 }
 516         }
 517         pipeunlock(rpipe);
 518
 519         if (error == 0)
 520                 vfs_timestamp(&rpipe->pipe_atime);
 521 unlocked_error:
 522         --rpipe->pipe_busy;
 523
 524         /*
 525          * PIPE_WANT processing only makes sense if pipe_busy is 0.
 526          */
 527         if ((rpipe->pipe_busy == 0) && (rpipe->pipe_state & PIPE_WANT)) {
 528                 rpipe->pipe_state &= ~(PIPE_WANT|PIPE_WANTW);
 529                 wakeup(rpipe);
 530         } else if (rpipe->pipe_buffer.cnt < MINPIPESIZE) {
 531                 /*
 532                  * Handle write blocking hysteresis.
 533                  */
 534                 if (rpipe->pipe_state & PIPE_WANTW) {
 535                         rpipe->pipe_state &= ~PIPE_WANTW;
 536                         wakeup(rpipe);
 537                 }
 538         }
 539
 540         if ((rpipe->pipe_buffer.size - rpipe->pipe_buffer.cnt) >= PIPE_BUF)
 541                 pipeselwakeup(rpipe);
 542
 543         return (error);
 544 }
 545
 546 #ifndef PIPE_NODIRECT
 547 /*
 548  * Map the sending processes' buffer into kernel space and wire it.
 549  * This is similar to a physical write operation.
 550  */
 551 static int
 552 pipe_build_write_buffer(wpipe, uio)
 553         struct pipe *wpipe;
 554         struct uio *uio;
 555 {
 556         u_int size;
 557         int i;
 558         vm_offset_t addr, endaddr;
 559         vm_paddr_t paddr;
 560
 561         size = (u_int) uio->uio_iov->iov_len;
 562         if (size > wpipe->pipe_buffer.size)
 563                 size = wpipe->pipe_buffer.size;
 564
 565         endaddr = round_page((vm_offset_t)uio->uio_iov->iov_base + size);
 566         addr = trunc_page((vm_offset_t)uio->uio_iov->iov_base);
 567         for (i = 0; addr < endaddr; addr += PAGE_SIZE, i++) {
 568                 vm_page_t m;
 569
 570                 if (vm_fault_quick((caddr_t)addr, VM_PROT_READ) < 0 ||
 571                     (paddr = pmap_kextract(addr)) == 0) {
 572                         int j;
 573
 574                         for (j = 0; j < i; j++)
 575                                 vm_page_unhold(wpipe->pipe_map.ms[j]);
 576                         return (EFAULT);
 577                 }
 578
 579                 m = PHYS_TO_VM_PAGE(paddr);
 580                 vm_page_hold(m);
 581                 wpipe->pipe_map.ms[i] = m;
 582         }
 583
 584 /*
 585  * set up the control block
 586  */
 587         wpipe->pipe_map.npages = i;
 588         wpipe->pipe_map.pos =
 589             ((vm_offset_t) uio->uio_iov->iov_base) & PAGE_MASK;
 590         wpipe->pipe_map.cnt = size;
 591
 592 /*
 593  * and map the buffer
 594  */
 595         if (wpipe->pipe_map.kva == 0) {
 596                 /*
 597                  * We need to allocate space for an extra page because the
 598                  * address range might (will) span pages at times.
 599                  */
 600                 wpipe->pipe_map.kva = kmem_alloc_pageable(kernel_map,
 601                         wpipe->pipe_buffer.size + PAGE_SIZE);
 602                 pipe_kva += wpipe->pipe_buffer.size + PAGE_SIZE;
 603                 ++pipe_dkmem_alloc;
 604         }
 605         pmap_qenter(wpipe->pipe_map.kva, wpipe->pipe_map.ms,
 606                 wpipe->pipe_map.npages);
 607
 608 /*
 609  * and update the uio data
 610  */
 611
 612         uio->uio_iov->iov_len -= size;
 613         uio->uio_iov->iov_base += size;
 614         if (uio->uio_iov->iov_len == 0)
 615                 uio->uio_iov++;
 616         uio->uio_resid -= size;
 617         uio->uio_offset += size;
 618         return (0);
 619 }
 620
 621 /*
 622  * unmap and unwire the process buffer
 623  */
 624 static void
 625 pipe_destroy_write_buffer(wpipe)
 626         struct pipe *wpipe;
 627 {
 628         int i;
 629
 630         if (wpipe->pipe_map.kva) {
 631                 pmap_qremove(wpipe->pipe_map.kva, wpipe->pipe_map.npages);
 632
 633                 if (pipe_kva > MAXPIPEKVA) {
 634                         vm_offset_t kva = wpipe->pipe_map.kva;
 635                         wpipe->pipe_map.kva = 0;
 636                         kmem_free(kernel_map, kva,
 637                                 wpipe->pipe_buffer.size + PAGE_SIZE);
 638                         pipe_kva -= wpipe->pipe_buffer.size + PAGE_SIZE;
 639                 }
 640         }
 641         for (i = 0; i < wpipe->pipe_map.npages; i++)
 642                 vm_page_unhold(wpipe->pipe_map.ms[i]);
 643         wpipe->pipe_map.npages = 0;
 644 }
 645
 646 /*
 647  * In the case of a signal, the writing process might go away.  This
 648  * code copies the data into the circular buffer so that the source
 649  * pages can be freed without loss of data.
 650  */
 651 static void
 652 pipe_clone_write_buffer(wpipe)
 653         struct pipe *wpipe;
 654 {
 655         int size;
 656         int pos;
 657
 658         size = wpipe->pipe_map.cnt;
 659         pos = wpipe->pipe_map.pos;
 660         bcopy((caddr_t) wpipe->pipe_map.kva + pos,
 661             (caddr_t) wpipe->pipe_buffer.buffer, size);
 662
 663         wpipe->pipe_buffer.in = size;
 664         wpipe->pipe_buffer.out = 0;
 665         wpipe->pipe_buffer.cnt = size;
 666         wpipe->pipe_state &= ~PIPE_DIRECTW;
 667
 668         pipe_destroy_write_buffer(wpipe);
 669 }
 670
 671 /*
 672  * This implements the pipe buffer write mechanism.  Note that only
 673  * a direct write OR a normal pipe write can be pending at any given time.
 674  * If there are any characters in the pipe buffer, the direct write will
 675  * be deferred until the receiving process grabs all of the bytes from
 676  * the pipe buffer.  Then the direct mapping write is set-up.
 677  */
 678 static int
 679 pipe_direct_write(wpipe, uio)
 680         struct pipe *wpipe;
 681         struct uio *uio;
 682 {
 683         int error;
 684
 685 retry:
 686         while (wpipe->pipe_state & PIPE_DIRECTW) {
 687                 if (wpipe->pipe_state & PIPE_WANTR) {
 688                         wpipe->pipe_state &= ~PIPE_WANTR;
 689                         wakeup(wpipe);
 690                 }
 691                 wpipe->pipe_state |= PIPE_WANTW;
 692                 error = tsleep(wpipe, PCATCH, "pipdww", 0);
 693                 if (error)
 694                         goto error1;
 695                 if (wpipe->pipe_state & PIPE_EOF) {
 696                         error = EPIPE;
 697                         goto error1;
 698                 }
 699         }
 700         wpipe->pipe_map.cnt = 0;        /* transfer not ready yet */
 701         if (wpipe->pipe_buffer.cnt > 0) {
 702                 if (wpipe->pipe_state & PIPE_WANTR) {
 703                         wpipe->pipe_state &= ~PIPE_WANTR;
 704                         wakeup(wpipe);
 705                 }
 706
 707                 wpipe->pipe_state |= PIPE_WANTW;
 708                 error = tsleep(wpipe, PCATCH, "pipdwc", 0);
 709                 if (error)
 710                         goto error1;
 711                 if (wpipe->pipe_state & PIPE_EOF) {
 712                         error = EPIPE;
 713                         goto error1;
 714                 }
 715                 goto retry;
 716         }
 717
 718         wpipe->pipe_state |= PIPE_DIRECTW;
 719
 720         error = pipe_build_write_buffer(wpipe, uio);
 721         if (error) {
 722                 wpipe->pipe_state &= ~PIPE_DIRECTW;
 723                 goto error1;
 724         }
 725
 726         error = 0;
 727         while (!error && (wpipe->pipe_state & PIPE_DIRECTW)) {
 728                 if (wpipe->pipe_state & PIPE_EOF) {
 729                         pipelock(wpipe, 0);
 730                         pipe_destroy_write_buffer(wpipe);
 731                         pipeunlock(wpipe);
 732                         pipeselwakeup(wpipe);
 733                         error = EPIPE;
 734                         goto error1;
 735                 }
 736                 if (wpipe->pipe_state & PIPE_WANTR) {
 737                         wpipe->pipe_state &= ~PIPE_WANTR;
 738                         wakeup(wpipe);
 739                 }
 740                 pipeselwakeup(wpipe);
 741                 error = tsleep(wpipe, PCATCH, "pipdwt", 0);
 742         }
 743
 744         pipelock(wpipe,0);
 745         if (wpipe->pipe_state & PIPE_DIRECTW) {
 746                 /*
 747                  * this bit of trickery substitutes a kernel buffer for
 748                  * the process that might be going away.
 749                  */
 750                 pipe_clone_write_buffer(wpipe);
 751         } else {
 752                 pipe_destroy_write_buffer(wpipe);
 753         }
 754         pipeunlock(wpipe);
 755         return (error);
 756
 757 error1:
 758         wakeup(wpipe);
 759         return (error);
 760 }
 761 #endif
 762
 763 static int
 764 pipe_write(struct file *fp, struct uio *uio, struct ucred *cred,
 765         int flags, struct thread *td)
 766 {
 767         int error = 0;
 768         int orig_resid;
 769         struct pipe *wpipe, *rpipe;
 770
 771         rpipe = (struct pipe *) fp->f_data;
 772         wpipe = rpipe->pipe_peer;
 773
 774         /*
 775          * detect loss of pipe read side, issue SIGPIPE if lost.
 776          */
 777         if ((wpipe == NULL) || (wpipe->pipe_state & PIPE_EOF)) {
 778                 return (EPIPE);
 779         }
 780         ++wpipe->pipe_busy;
 781
 782         /*
 783          * If it is advantageous to resize the pipe buffer, do
 784          * so.
 785          */
 786         if ((uio->uio_resid > PIPE_SIZE) &&
 787                 (pipe_nbig < pipe_maxbig) &&
 788                 (wpipe->pipe_state & PIPE_DIRECTW) == 0 &&
 789                 (wpipe->pipe_buffer.size <= PIPE_SIZE) &&
 790                 (wpipe->pipe_buffer.cnt == 0)) {
 791
 792                 if ((error = pipelock(wpipe,1)) == 0) {
 793                         if (pipespace(wpipe, BIG_PIPE_SIZE) == 0)
 794                                 pipe_nbig++;
 795                         pipeunlock(wpipe);
 796                 }
 797         }
 798
 799         /*
 800          * If an early error occured unbusy and return, waking up any pending
 801          * readers.
 802          */
 803         if (error) {
 804                 --wpipe->pipe_busy;
 805                 if ((wpipe->pipe_busy == 0) &&
 806                     (wpipe->pipe_state & PIPE_WANT)) {
 807                         wpipe->pipe_state &= ~(PIPE_WANT | PIPE_WANTR);
 808                         wakeup(wpipe);
 809                 }
 810                 return(error);
 811         }
 812
 813         KASSERT(wpipe->pipe_buffer.buffer != NULL, ("pipe buffer gone"));
 814
 815         orig_resid = uio->uio_resid;
 816
 817         while (uio->uio_resid) {
 818                 int space;
 819
 820 #ifndef PIPE_NODIRECT
 821                 /*
 822                  * If the transfer is large, we can gain performance if
 823                  * we do process-to-process copies directly.
 824                  * If the write is non-blocking, we don't use the
 825                  * direct write mechanism.
 826                  *
 827                  * The direct write mechanism will detect the reader going
 828                  * away on us.
 829                  */
 830                 if ((uio->uio_iov->iov_len >= PIPE_MINDIRECT) &&
 831                     (fp->f_flag & FNONBLOCK) == 0 &&
 832                         (wpipe->pipe_map.kva || (pipe_kva < LIMITPIPEKVA)) &&
 833                         (uio->uio_iov->iov_len >= PIPE_MINDIRECT)) {
 834                         error = pipe_direct_write( wpipe, uio);
 835                         if (error)
 836                                 break;
 837                         continue;
 838                 }
 839 #endif
 840
 841                 /*
 842                  * Pipe buffered writes cannot be coincidental with
 843                  * direct writes.  We wait until the currently executing
 844                  * direct write is completed before we start filling the
 845                  * pipe buffer.  We break out if a signal occurs or the
 846                  * reader goes away.
 847                  */
 848         retrywrite:
 849                 while (wpipe->pipe_state & PIPE_DIRECTW) {
 850                         if (wpipe->pipe_state & PIPE_WANTR) {
 851                                 wpipe->pipe_state &= ~PIPE_WANTR;
 852                                 wakeup(wpipe);
 853                         }
 854                         error = tsleep(wpipe, PCATCH, "pipbww", 0);
 855                         if (wpipe->pipe_state & PIPE_EOF)
 856                                 break;
 857                         if (error)
 858                                 break;
 859                 }
 860                 if (wpipe->pipe_state & PIPE_EOF) {
 861                         error = EPIPE;
 862                         break;
 863                 }
 864
 865                 space = wpipe->pipe_buffer.size - wpipe->pipe_buffer.cnt;
 866
 867                 /* Writes of size <= PIPE_BUF must be atomic. */
 868                 if ((space < uio->uio_resid) && (orig_resid <= PIPE_BUF))
 869                         space = 0;
 870
 871                 if (space > 0 && (wpipe->pipe_buffer.cnt < PIPE_SIZE)) {
 872                         if ((error = pipelock(wpipe,1)) == 0) {
 873                                 int size;       /* Transfer size */
 874                                 int segsize;    /* first segment to transfer */
 875
 876                                 /*
 877                                  * It is possible for a direct write to
 878                                  * slip in on us... handle it here...
 879                                  */
 880                                 if (wpipe->pipe_state & PIPE_DIRECTW) {
 881                                         pipeunlock(wpipe);
 882                                         goto retrywrite;
 883                                 }
 884                                 /*
 885                                  * If a process blocked in uiomove, our
 886                                  * value for space might be bad.
 887                                  *
 888                                  * XXX will we be ok if the reader has gone
 889                                  * away here?
 890                                  */
 891                                 if (space > wpipe->pipe_buffer.size -
 892                                     wpipe->pipe_buffer.cnt) {
 893                                         pipeunlock(wpipe);
 894                                         goto retrywrite;
 895                                 }
 896
 897                                 /*
 898                                  * Transfer size is minimum of uio transfer
 899                                  * and free space in pipe buffer.
 900                                  */
 901                                 if (space > uio->uio_resid)
 902                                         size = uio->uio_resid;
 903                                 else
 904                                         size = space;
 905                                 /*
 906                                  * First segment to transfer is minimum of
 907                                  * transfer size and contiguous space in
 908                                  * pipe buffer.  If first segment to transfer
 909                                  * is less than the transfer size, we've got
 910                                  * a wraparound in the buffer.
 911                                  */
 912                                 segsize = wpipe->pipe_buffer.size -
 913                                         wpipe->pipe_buffer.in;
 914                                 if (segsize > size)
 915                                         segsize = size;
 916
 917                                 /* Transfer first segment */
 918
 919                                 error = uiomove(&wpipe->pipe_buffer.buffer[wpipe->pipe_buffer.in],
 920                                                 segsize, uio);
 921
 922                                 if (error == 0 && segsize < size) {
 923                                         /*
 924                                          * Transfer remaining part now, to
 925                                          * support atomic writes.  Wraparound
 926                                          * happened.
 927                                          */
 928                                         if (wpipe->pipe_buffer.in + segsize !=
 929                                             wpipe->pipe_buffer.size)
 930                                                 panic("Expected pipe buffer wraparound disappeared");
 931
 932                                         error = uiomove(&wpipe->pipe_buffer.buffer[0],
 933                                                         size - segsize, uio);
 934                                 }
 935                                 if (error == 0) {
 936                                         wpipe->pipe_buffer.in += size;
 937                                         if (wpipe->pipe_buffer.in >=
 938                                             wpipe->pipe_buffer.size) {
 939                                                 if (wpipe->pipe_buffer.in != size - segsize + wpipe->pipe_buffer.size)
 940                                                         panic("Expected wraparound bad");
 941                                                 wpipe->pipe_buffer.in = size - segsize;
 942                                         }
 943
 944                                         wpipe->pipe_buffer.cnt += size;
 945                                         if (wpipe->pipe_buffer.cnt > wpipe->pipe_buffer.size)
 946                                                 panic("Pipe buffer overflow");
 947
 948                                 }
 949                                 pipeunlock(wpipe);
 950                         }
 951                         if (error)
 952                                 break;
 953
 954                 } else {
 955                         /*
 956                          * If the "read-side" has been blocked, wake it up now.
 957                          */
 958                         if (wpipe->pipe_state & PIPE_WANTR) {
 959                                 wpipe->pipe_state &= ~PIPE_WANTR;
 960                                 wakeup(wpipe);
 961                         }
 962
 963                         /*
 964                          * don't block on non-blocking I/O
 965                          */
 966                         if (fp->f_flag & FNONBLOCK) {
 967                                 error = EAGAIN;
 968                                 break;
 969                         }
 970
 971                         /*
 972                          * We have no more space and have something to offer,
 973                          * wake up select/poll.
 974                          */
 975                         pipeselwakeup(wpipe);
 976
 977                         wpipe->pipe_state |= PIPE_WANTW;
 978                         error = tsleep(wpipe, PCATCH, "pipewr", 0);
 979                         if (error != 0)
 980                                 break;
 981                         /*
 982                          * If read side wants to go away, we just issue a signal
 983                          * to ourselves.
 984                          */
 985                         if (wpipe->pipe_state & PIPE_EOF) {
 986                                 error = EPIPE;
 987                                 break;
 988                         }
 989                 }
 990         }
 991
 992         --wpipe->pipe_busy;
 993
 994         if ((wpipe->pipe_busy == 0) && (wpipe->pipe_state & PIPE_WANT)) {
 995                 wpipe->pipe_state &= ~(PIPE_WANT | PIPE_WANTR);
 996                 wakeup(wpipe);
 997         } else if (wpipe->pipe_buffer.cnt > 0) {
 998                 /*
 999                  * If we have put any characters in the buffer, we wake up
1000                  * the reader.
1001                  */
1002                 if (wpipe->pipe_state & PIPE_WANTR) {
1003                         wpipe->pipe_state &= ~PIPE_WANTR;
1004                         wakeup(wpipe);
1005                 }
1006         }
1007
1008         /*
1009          * Don't return EPIPE if I/O was successful
1010          */
1011         if ((wpipe->pipe_buffer.cnt == 0) &&
1012             (uio->uio_resid == 0) &&
1013             (error == EPIPE)) {
1014                 error = 0;
1015         }
1016
1017         if (error == 0)
1018                 vfs_timestamp(&wpipe->pipe_mtime);
1019
1020         /*
1021          * We have something to offer,
1022          * wake up select/poll.
1023          */
1024         if (wpipe->pipe_buffer.cnt)
1025                 pipeselwakeup(wpipe);
1026
1027         return (error);
1028 }
1029
1030 /*
1031  * we implement a very minimal set of ioctls for compatibility with sockets.
1032  */
1033 int
1034 pipe_ioctl(struct file *fp, u_long cmd, caddr_t data, struct thread *td)
1035 {
1036         struct pipe *mpipe = (struct pipe *)fp->f_data;
1037
1038         switch (cmd) {
1039
1040         case FIONBIO:
1041                 return (0);
1042
1043         case FIOASYNC:
1044                 if (*(int *)data) {
1045                         mpipe->pipe_state |= PIPE_ASYNC;
1046                 } else {
1047                         mpipe->pipe_state &= ~PIPE_ASYNC;
1048                 }
1049                 return (0);
1050
1051         case FIONREAD:
1052                 if (mpipe->pipe_state & PIPE_DIRECTW)
1053                         *(int *)data = mpipe->pipe_map.cnt;
1054                 else
1055                         *(int *)data = mpipe->pipe_buffer.cnt;
1056                 return (0);
1057
1058         case FIOSETOWN:
1059                 return (fsetown(*(int *)data, &mpipe->pipe_sigio));
1060
1061         case FIOGETOWN:
1062                 *(int *)data = fgetown(mpipe->pipe_sigio);
1063                 return (0);
1064
1065         /* This is deprecated, FIOSETOWN should be used instead. */
1066         case TIOCSPGRP:
1067                 return (fsetown(-(*(int *)data), &mpipe->pipe_sigio));
1068
1069         /* This is deprecated, FIOGETOWN should be used instead. */
1070         case TIOCGPGRP:
1071                 *(int *)data = -fgetown(mpipe->pipe_sigio);
1072                 return (0);
1073
1074         }
1075         return (ENOTTY);
1076 }
1077
1078 int
1079 pipe_poll(struct file *fp, int events, struct ucred *cred, struct thread *td)
1080 {
1081         struct pipe *rpipe = (struct pipe *)fp->f_data;
1082         struct pipe *wpipe;
1083         int revents = 0;
1084
1085         wpipe = rpipe->pipe_peer;
1086         if (events & (POLLIN | POLLRDNORM))
1087                 if ((rpipe->pipe_state & PIPE_DIRECTW) ||
1088                     (rpipe->pipe_buffer.cnt > 0) ||
1089                     (rpipe->pipe_state & PIPE_EOF))
1090                         revents |= events & (POLLIN | POLLRDNORM);
1091
1092         if (events & (POLLOUT | POLLWRNORM))
1093                 if (wpipe == NULL || (wpipe->pipe_state & PIPE_EOF) ||
1094                     (((wpipe->pipe_state & PIPE_DIRECTW) == 0) &&
1095                      (wpipe->pipe_buffer.size - wpipe->pipe_buffer.cnt) >= PIPE_BUF))
1096                         revents |= events & (POLLOUT | POLLWRNORM);
1097
1098         if ((rpipe->pipe_state & PIPE_EOF) ||
1099             (wpipe == NULL) ||
1100             (wpipe->pipe_state & PIPE_EOF))
1101                 revents |= POLLHUP;
1102
1103         if (revents == 0) {
1104                 if (events & (POLLIN | POLLRDNORM)) {
1105                         selrecord(td, &rpipe->pipe_sel);
1106                         rpipe->pipe_state |= PIPE_SEL;
1107                 }
1108
1109                 if (events & (POLLOUT | POLLWRNORM)) {
1110                         selrecord(td, &wpipe->pipe_sel);
1111                         wpipe->pipe_state |= PIPE_SEL;
1112                 }
1113         }
1114
1115         return (revents);
1116 }
1117
1118 static int
1119 pipe_stat(struct file *fp, struct stat *ub, struct thread *td)
1120 {
1121         struct pipe *pipe = (struct pipe *)fp->f_data;
1122
1123         bzero((caddr_t)ub, sizeof(*ub));
1124         ub->st_mode = S_IFIFO;
1125         ub->st_blksize = pipe->pipe_buffer.size;
1126         ub->st_size = pipe->pipe_buffer.cnt;
1127         ub->st_blocks = (ub->st_size + ub->st_blksize - 1) / ub->st_blksize;
1128         ub->st_atimespec = pipe->pipe_atime;
1129         ub->st_mtimespec = pipe->pipe_mtime;
1130         ub->st_ctimespec = pipe->pipe_ctime;
1131         /*
1132          * Left as 0: st_dev, st_ino, st_nlink, st_uid, st_gid, st_rdev,
1133          * st_flags, st_gen.
1134          * XXX (st_dev, st_ino) should be unique.
1135          */
1136         return (0);
1137 }
1138
1139 /* ARGSUSED */
1140 static int
1141 pipe_close(struct file *fp, struct thread *td)
1142 {
1143         struct pipe *cpipe = (struct pipe *)fp->f_data;
1144
1145         fp->f_ops = &badfileops;
1146         fp->f_data = NULL;
1147         funsetown(cpipe->pipe_sigio);
1148         pipeclose(cpipe);
1149         return (0);
1150 }
1151
1152 static void
1153 pipe_free_kmem(struct pipe *cpipe)
1154 {
1155         if (cpipe->pipe_buffer.buffer != NULL) {
1156                 if (cpipe->pipe_buffer.size > PIPE_SIZE)
1157                         --pipe_nbig;
1158                 pipe_kva -= cpipe->pipe_buffer.size;
1159                 kmem_free(kernel_map,
1160                         (vm_offset_t)cpipe->pipe_buffer.buffer,
1161                         cpipe->pipe_buffer.size);
1162                 cpipe->pipe_buffer.buffer = NULL;
1163                 cpipe->pipe_buffer.object = NULL;
1164         }
1165 #ifndef PIPE_NODIRECT
1166         if (cpipe->pipe_map.kva != NULL) {
1167                 pipe_kva -= cpipe->pipe_buffer.size + PAGE_SIZE;
1168                 kmem_free(kernel_map,
1169                         cpipe->pipe_map.kva,
1170                         cpipe->pipe_buffer.size + PAGE_SIZE);
1171                 cpipe->pipe_map.cnt = 0;
1172                 cpipe->pipe_map.kva = 0;
1173                 cpipe->pipe_map.pos = 0;
1174                 cpipe->pipe_map.npages = 0;
1175         }
1176 #endif
1177 }
1178
1179 /*
1180  * shutdown the pipe
1181  */
1182 static void
1183 pipeclose(struct pipe *cpipe)
1184 {
1185         globaldata_t gd;
1186         struct pipe *ppipe;
1187
1188         if (cpipe == NULL)
1189                 return;
1190
1191         pipeselwakeup(cpipe);
1192
1193         /*
1194          * If the other side is blocked, wake it up saying that
1195          * we want to close it down.
1196          */
1197         while (cpipe->pipe_busy) {
1198                 wakeup(cpipe);
1199                 cpipe->pipe_state |= PIPE_WANT | PIPE_EOF;
1200                 tsleep(cpipe, 0, "pipecl", 0);
1201         }
1202
1203         /*
1204          * Disconnect from peer
1205          */
1206         if ((ppipe = cpipe->pipe_peer) != NULL) {
1207                 pipeselwakeup(ppipe);
1208
1209                 ppipe->pipe_state |= PIPE_EOF;
1210                 wakeup(ppipe);
1211                 KNOTE(&ppipe->pipe_sel.si_note, 0);
1212                 ppipe->pipe_peer = NULL;
1213         }
1214
1215         /*
1216          * free or cache resources
1217          */
1218         gd = mycpu;
1219         if (gd->gd_pipeqcount >= pipe_maxcache ||
1220             cpipe->pipe_buffer.size != PIPE_SIZE
1221         ) {
1222                 pipe_free_kmem(cpipe);
1223                 free(cpipe, M_PIPE);
1224         } else {
1225                 KKASSERT(cpipe->pipe_map.npages == 0);
1226
1227                 cpipe->pipe_state = 0;
1228                 cpipe->pipe_busy = 0;
1229                 cpipe->pipe_map.cnt = 0;
1230                 cpipe->pipe_map.pos = 0;
1231                 cpipe->pipe_peer = gd->gd_pipeq;
1232                 gd->gd_pipeq = cpipe;
1233                 ++gd->gd_pipeqcount;
1234         }
1235 }
1236
1237 /*ARGSUSED*/
1238 static int
1239 pipe_kqfilter(struct file *fp, struct knote *kn)
1240 {
1241         struct pipe *cpipe = (struct pipe *)kn->kn_fp->f_data;
1242
1243         switch (kn->kn_filter) {
1244         case EVFILT_READ:
1245                 kn->kn_fop = &pipe_rfiltops;
1246                 break;
1247         case EVFILT_WRITE:
1248                 kn->kn_fop = &pipe_wfiltops;
1249                 cpipe = cpipe->pipe_peer;
1250                 if (cpipe == NULL)
1251                         /* other end of pipe has been closed */
1252                         return (EPIPE);
1253                 break;
1254         default:
1255                 return (1);
1256         }
1257         kn->kn_hook = (caddr_t)cpipe;
1258
1259         SLIST_INSERT_HEAD(&cpipe->pipe_sel.si_note, kn, kn_selnext);
1260         return (0);
1261 }
1262
1263 static void
1264 filt_pipedetach(struct knote *kn)
1265 {
1266         struct pipe *cpipe = (struct pipe *)kn->kn_hook;
1267
1268         SLIST_REMOVE(&cpipe->pipe_sel.si_note, kn, knote, kn_selnext);
1269 }
1270
1271 /*ARGSUSED*/
1272 static int
1273 filt_piperead(struct knote *kn, long hint)
1274 {
1275         struct pipe *rpipe = (struct pipe *)kn->kn_fp->f_data;
1276         struct pipe *wpipe = rpipe->pipe_peer;
1277
1278         kn->kn_data = rpipe->pipe_buffer.cnt;
1279         if ((kn->kn_data == 0) && (rpipe->pipe_state & PIPE_DIRECTW))
1280                 kn->kn_data = rpipe->pipe_map.cnt;
1281
1282         if ((rpipe->pipe_state & PIPE_EOF) ||
1283             (wpipe == NULL) || (wpipe->pipe_state & PIPE_EOF)) {
1284                 kn->kn_flags |= EV_EOF;
1285                 return (1);
1286         }
1287         return (kn->kn_data > 0);
1288 }
1289
1290 /*ARGSUSED*/
1291 static int
1292 filt_pipewrite(struct knote *kn, long hint)
1293 {
1294         struct pipe *rpipe = (struct pipe *)kn->kn_fp->f_data;
1295         struct pipe *wpipe = rpipe->pipe_peer;
1296
1297         if ((wpipe == NULL) || (wpipe->pipe_state & PIPE_EOF)) {
1298                 kn->kn_data = 0;
1299                 kn->kn_flags |= EV_EOF;
1300                 return (1);
1301         }
1302         kn->kn_data = wpipe->pipe_buffer.size - wpipe->pipe_buffer.cnt;
1303         if (wpipe->pipe_state & PIPE_DIRECTW)
1304                 kn->kn_data = 0;
1305
1306         return (kn->kn_data >= PIPE_BUF);
1307 }