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 1.50 2008/09/09 04:06:13 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 #include <sys/param.h>
  30 #include <sys/systm.h>
  31 #include <sys/kernel.h>
  32 #include <sys/proc.h>
  33 #include <sys/fcntl.h>
  34 #include <sys/file.h>
  35 #include <sys/filedesc.h>
  36 #include <sys/filio.h>
  37 #include <sys/ttycom.h>
  38 #include <sys/stat.h>
  39 #include <sys/signalvar.h>
  40 #include <sys/sysproto.h>
  41 #include <sys/pipe.h>
  42 #include <sys/vnode.h>
  43 #include <sys/uio.h>
  44 #include <sys/event.h>
  45 #include <sys/globaldata.h>
  46 #include <sys/module.h>
  47 #include <sys/malloc.h>
  48 #include <sys/sysctl.h>
  49 #include <sys/socket.h>
  50
  51 #include <vm/vm.h>
  52 #include <vm/vm_param.h>
  53 #include <sys/lock.h>
  54 #include <vm/vm_object.h>
  55 #include <vm/vm_kern.h>
  56 #include <vm/vm_extern.h>
  57 #include <vm/pmap.h>
  58 #include <vm/vm_map.h>
  59 #include <vm/vm_page.h>
  60 #include <vm/vm_zone.h>
  61
  62 #include <sys/file2.h>
  63 #include <sys/signal2.h>
  64
  65 #include <machine/cpufunc.h>
  66
  67 /*
  68  * interfaces to the outside world
  69  */
  70 static int pipe_read (struct file *fp, struct uio *uio,
  71                 struct ucred *cred, int flags);
  72 static int pipe_write (struct file *fp, struct uio *uio,
  73                 struct ucred *cred, int flags);
  74 static int pipe_close (struct file *fp);
  75 static int pipe_shutdown (struct file *fp, int how);
  76 static int pipe_kqfilter (struct file *fp, struct knote *kn);
  77 static int pipe_stat (struct file *fp, struct stat *sb, struct ucred *cred);
  78 static int pipe_ioctl (struct file *fp, u_long cmd, caddr_t data,
  79                 struct ucred *cred, struct sysmsg *msg);
  80
  81 static struct fileops pipeops = {
  82         .fo_read = pipe_read,
  83         .fo_write = pipe_write,
  84         .fo_ioctl = pipe_ioctl,
  85         .fo_kqfilter = pipe_kqfilter,
  86         .fo_stat = pipe_stat,
  87         .fo_close = pipe_close,
  88         .fo_shutdown = pipe_shutdown
  89 };
  90
  91 static void     filt_pipedetach(struct knote *kn);
  92 static int      filt_piperead(struct knote *kn, long hint);
  93 static int      filt_pipewrite(struct knote *kn, long hint);
  94
  95 static struct filterops pipe_rfiltops =
  96         { FILTEROP_ISFD|FILTEROP_MPSAFE, NULL, filt_pipedetach, filt_piperead };
  97 static struct filterops pipe_wfiltops =
  98         { FILTEROP_ISFD|FILTEROP_MPSAFE, NULL, filt_pipedetach, filt_pipewrite };
  99
 100 MALLOC_DEFINE(M_PIPE, "pipe", "pipe structures");
 101
 102 /*
 103  * Default pipe buffer size(s), this can be kind-of large now because pipe
 104  * space is pageable.  The pipe code will try to maintain locality of
 105  * reference for performance reasons, so small amounts of outstanding I/O
 106  * will not wipe the cache.
 107  */
 108 #define MINPIPESIZE (PIPE_SIZE/3)
 109 #define MAXPIPESIZE (2*PIPE_SIZE/3)
 110
 111 /*
 112  * Limit the number of "big" pipes
 113  */
 114 #define LIMITBIGPIPES   64
 115 #define PIPEQ_MAX_CACHE 16      /* per-cpu pipe structure cache */
 116
 117 static int pipe_maxbig = LIMITBIGPIPES;
 118 static int pipe_maxcache = PIPEQ_MAX_CACHE;
 119 static int pipe_bigcount;
 120 static int pipe_nbig;
 121 static int pipe_bcache_alloc;
 122 static int pipe_bkmem_alloc;
 123 static int pipe_rblocked_count;
 124 static int pipe_wblocked_count;
 125
 126 SYSCTL_NODE(_kern, OID_AUTO, pipe, CTLFLAG_RW, 0, "Pipe operation");
 127 SYSCTL_INT(_kern_pipe, OID_AUTO, nbig,
 128         CTLFLAG_RD, &pipe_nbig, 0, "numer of big pipes allocated");
 129 SYSCTL_INT(_kern_pipe, OID_AUTO, bigcount,
 130         CTLFLAG_RW, &pipe_bigcount, 0, "number of times pipe expanded");
 131 SYSCTL_INT(_kern_pipe, OID_AUTO, rblocked,
 132         CTLFLAG_RW, &pipe_rblocked_count, 0, "number of times pipe expanded");
 133 SYSCTL_INT(_kern_pipe, OID_AUTO, wblocked,
 134         CTLFLAG_RW, &pipe_wblocked_count, 0, "number of times pipe expanded");
 135 SYSCTL_INT(_kern_pipe, OID_AUTO, maxcache,
 136         CTLFLAG_RW, &pipe_maxcache, 0, "max pipes cached per-cpu");
 137 SYSCTL_INT(_kern_pipe, OID_AUTO, maxbig,
 138         CTLFLAG_RW, &pipe_maxbig, 0, "max number of big pipes");
 139 #ifdef SMP
 140 static int pipe_delay = 5000;   /* 5uS default */
 141 SYSCTL_INT(_kern_pipe, OID_AUTO, delay,
 142         CTLFLAG_RW, &pipe_delay, 0, "SMP delay optimization in ns");
 143 #endif
 144 #if !defined(NO_PIPE_SYSCTL_STATS)
 145 SYSCTL_INT(_kern_pipe, OID_AUTO, bcache_alloc,
 146         CTLFLAG_RW, &pipe_bcache_alloc, 0, "pipe buffer from pcpu cache");
 147 SYSCTL_INT(_kern_pipe, OID_AUTO, bkmem_alloc,
 148         CTLFLAG_RW, &pipe_bkmem_alloc, 0, "pipe buffer from kmem");
 149 #endif
 150
 151 static void pipeclose (struct pipe *cpipe);
 152 static void pipe_free_kmem (struct pipe *cpipe);
 153 static int pipe_create (struct pipe **cpipep);
 154 static int pipespace (struct pipe *cpipe, int size);
 155
 156 static __inline void
 157 pipewakeup(struct pipe *cpipe, int dosigio)
 158 {
 159         if (dosigio && (cpipe->pipe_state & PIPE_ASYNC) && cpipe->pipe_sigio) {
 160                 lwkt_gettoken(&proc_token);
 161                 pgsigio(cpipe->pipe_sigio, SIGIO, 0);
 162                 lwkt_reltoken(&proc_token);
 163         }
 164         KNOTE(&cpipe->pipe_kq.ki_note, 0);
 165 }
 166
 167 /*
 168  * These routines are called before and after a UIO.  The UIO
 169  * may block, causing our held tokens to be lost temporarily.
 170  *
 171  * We use these routines to serialize reads against other reads
 172  * and writes against other writes.
 173  *
 174  * The read token is held on entry so *ipp does not race.
 175  */
 176 static __inline int
 177 pipe_start_uio(struct pipe *cpipe, int *ipp)
 178 {
 179         int error;
 180
 181         while (*ipp) {
 182                 *ipp = -1;
 183                 error = tsleep(ipp, PCATCH, "pipexx", 0);
 184                 if (error)
 185                         return (error);
 186         }
 187         *ipp = 1;
 188         return (0);
 189 }
 190
 191 static __inline void
 192 pipe_end_uio(struct pipe *cpipe, int *ipp)
 193 {
 194         if (*ipp < 0) {
 195                 *ipp = 0;
 196                 wakeup(ipp);
 197         } else {
 198                 KKASSERT(*ipp > 0);
 199                 *ipp = 0;
 200         }
 201 }
 202
 203 /*
 204  * The pipe system call for the DTYPE_PIPE type of pipes
 205  *
 206  * pipe_args(int dummy)
 207  *
 208  * MPSAFE
 209  */
 210 int
 211 sys_pipe(struct pipe_args *uap)
 212 {
 213         struct thread *td = curthread;
 214         struct filedesc *fdp = td->td_proc->p_fd;
 215         struct file *rf, *wf;
 216         struct pipe *rpipe, *wpipe;
 217         int fd1, fd2, error;
 218
 219         rpipe = wpipe = NULL;
 220         if (pipe_create(&rpipe) || pipe_create(&wpipe)) {
 221                 pipeclose(rpipe);
 222                 pipeclose(wpipe);
 223                 return (ENFILE);
 224         }
 225
 226         error = falloc(td->td_lwp, &rf, &fd1);
 227         if (error) {
 228                 pipeclose(rpipe);
 229                 pipeclose(wpipe);
 230                 return (error);
 231         }
 232         uap->sysmsg_fds[0] = fd1;
 233
 234         /*
 235          * Warning: once we've gotten past allocation of the fd for the
 236          * read-side, we can only drop the read side via fdrop() in order
 237          * to avoid races against processes which manage to dup() the read
 238          * side while we are blocked trying to allocate the write side.
 239          */
 240         rf->f_type = DTYPE_PIPE;
 241         rf->f_flag = FREAD | FWRITE;
 242         rf->f_ops = &pipeops;
 243         rf->f_data = rpipe;
 244         error = falloc(td->td_lwp, &wf, &fd2);
 245         if (error) {
 246                 fsetfd(fdp, NULL, fd1);
 247                 fdrop(rf);
 248                 /* rpipe has been closed by fdrop(). */
 249                 pipeclose(wpipe);
 250                 return (error);
 251         }
 252         wf->f_type = DTYPE_PIPE;
 253         wf->f_flag = FREAD | FWRITE;
 254         wf->f_ops = &pipeops;
 255         wf->f_data = wpipe;
 256         uap->sysmsg_fds[1] = fd2;
 257
 258         rpipe->pipe_slock = kmalloc(sizeof(struct lock),
 259                                     M_PIPE, M_WAITOK|M_ZERO);
 260         wpipe->pipe_slock = rpipe->pipe_slock;
 261         rpipe->pipe_peer = wpipe;
 262         wpipe->pipe_peer = rpipe;
 263         lockinit(rpipe->pipe_slock, "pipecl", 0, 0);
 264
 265         /*
 266          * Once activated the peer relationship remains valid until
 267          * both sides are closed.
 268          */
 269         fsetfd(fdp, rf, fd1);
 270         fsetfd(fdp, wf, fd2);
 271         fdrop(rf);
 272         fdrop(wf);
 273
 274         return (0);
 275 }
 276
 277 /*
 278  * Allocate kva for pipe circular buffer, the space is pageable
 279  * This routine will 'realloc' the size of a pipe safely, if it fails
 280  * it will retain the old buffer.
 281  * If it fails it will return ENOMEM.
 282  */
 283 static int
 284 pipespace(struct pipe *cpipe, int size)
 285 {
 286         struct vm_object *object;
 287         caddr_t buffer;
 288         int npages, error;
 289
 290         npages = round_page(size) / PAGE_SIZE;
 291         object = cpipe->pipe_buffer.object;
 292
 293         /*
 294          * [re]create the object if necessary and reserve space for it
 295          * in the kernel_map.  The object and memory are pageable.  On
 296          * success, free the old resources before assigning the new
 297          * ones.
 298          */
 299         if (object == NULL || object->size != npages) {
 300                 object = vm_object_allocate(OBJT_DEFAULT, npages);
 301                 buffer = (caddr_t)vm_map_min(&kernel_map);
 302
 303                 error = vm_map_find(&kernel_map, object, 0,
 304                                     (vm_offset_t *)&buffer,
 305                                     size, PAGE_SIZE,
 306                                     1, VM_MAPTYPE_NORMAL,
 307                                     VM_PROT_ALL, VM_PROT_ALL,
 308                                     0);
 309
 310                 if (error != KERN_SUCCESS) {
 311                         vm_object_deallocate(object);
 312                         return (ENOMEM);
 313                 }
 314                 pipe_free_kmem(cpipe);
 315                 cpipe->pipe_buffer.object = object;
 316                 cpipe->pipe_buffer.buffer = buffer;
 317                 cpipe->pipe_buffer.size = size;
 318                 ++pipe_bkmem_alloc;
 319         } else {
 320                 ++pipe_bcache_alloc;
 321         }
 322         cpipe->pipe_buffer.rindex = 0;
 323         cpipe->pipe_buffer.windex = 0;
 324         return (0);
 325 }
 326
 327 /*
 328  * Initialize and allocate VM and memory for pipe, pulling the pipe from
 329  * our per-cpu cache if possible.  For now make sure it is sized for the
 330  * smaller PIPE_SIZE default.
 331  */
 332 static int
 333 pipe_create(struct pipe **cpipep)
 334 {
 335         globaldata_t gd = mycpu;
 336         struct pipe *cpipe;
 337         int error;
 338
 339         if ((cpipe = gd->gd_pipeq) != NULL) {
 340                 gd->gd_pipeq = cpipe->pipe_peer;
 341                 --gd->gd_pipeqcount;
 342                 cpipe->pipe_peer = NULL;
 343                 cpipe->pipe_wantwcnt = 0;
 344         } else {
 345                 cpipe = kmalloc(sizeof(struct pipe), M_PIPE, M_WAITOK|M_ZERO);
 346         }
 347         *cpipep = cpipe;
 348         if ((error = pipespace(cpipe, PIPE_SIZE)) != 0)
 349                 return (error);
 350         vfs_timestamp(&cpipe->pipe_ctime);
 351         cpipe->pipe_atime = cpipe->pipe_ctime;
 352         cpipe->pipe_mtime = cpipe->pipe_ctime;
 353         lwkt_token_init(&cpipe->pipe_rlock, "piper");
 354         lwkt_token_init(&cpipe->pipe_wlock, "pipew");
 355         return (0);
 356 }
 357
 358 static int
 359 pipe_read(struct file *fp, struct uio *uio, struct ucred *cred, int fflags)
 360 {
 361         struct pipe *rpipe;
 362         struct pipe *wpipe;
 363         int error;
 364         size_t nread = 0;
 365         int nbio;
 366         u_int size;     /* total bytes available */
 367         u_int nsize;    /* total bytes to read */
 368         u_int rindex;   /* contiguous bytes available */
 369         int notify_writer;
 370         int bigread;
 371         int bigcount;
 372
 373         if (uio->uio_resid == 0)
 374                 return(0);
 375
 376         /*
 377          * Setup locks, calculate nbio
 378          */
 379         rpipe = (struct pipe *)fp->f_data;
 380         wpipe = rpipe->pipe_peer;
 381         lwkt_gettoken(&rpipe->pipe_rlock);
 382
 383         if (fflags & O_FBLOCKING)
 384                 nbio = 0;
 385         else if (fflags & O_FNONBLOCKING)
 386                 nbio = 1;
 387         else if (fp->f_flag & O_NONBLOCK)
 388                 nbio = 1;
 389         else
 390                 nbio = 0;
 391
 392         /*
 393          * Reads are serialized.  Note however that pipe_buffer.buffer and
 394          * pipe_buffer.size can change out from under us when the number
 395          * of bytes in the buffer are zero due to the write-side doing a
 396          * pipespace().
 397          */
 398         error = pipe_start_uio(rpipe, &rpipe->pipe_rip);
 399         if (error) {
 400                 lwkt_reltoken(&rpipe->pipe_rlock);
 401                 return (error);
 402         }
 403         notify_writer = 0;
 404
 405         bigread = (uio->uio_resid > 10 * 1024 * 1024);
 406         bigcount = 10;
 407
 408         while (uio->uio_resid) {
 409                 /*
 410                  * Don't hog the cpu.
 411                  */
 412                 if (bigread && --bigcount == 0) {
 413                         lwkt_user_yield();
 414                         bigcount = 10;
 415                         if (CURSIG(curthread->td_lwp)) {
 416                                 error = EINTR;
 417                                 break;
 418                         }
 419                 }
 420
 421                 size = rpipe->pipe_buffer.windex - rpipe->pipe_buffer.rindex;
 422                 cpu_lfence();
 423                 if (size) {
 424                         rindex = rpipe->pipe_buffer.rindex &
 425                                  (rpipe->pipe_buffer.size - 1);
 426                         nsize = size;
 427                         if (nsize > rpipe->pipe_buffer.size - rindex)
 428                                 nsize = rpipe->pipe_buffer.size - rindex;
 429                         nsize = szmin(nsize, uio->uio_resid);
 430
 431                         error = uiomove(&rpipe->pipe_buffer.buffer[rindex],
 432                                         nsize, uio);
 433                         if (error)
 434                                 break;
 435                         cpu_mfence();
 436                         rpipe->pipe_buffer.rindex += nsize;
 437                         nread += nsize;
 438
 439                         /*
 440                          * If the FIFO is still over half full just continue
 441                          * and do not try to notify the writer yet.
 442                          */
 443                         if (size - nsize >= (rpipe->pipe_buffer.size >> 1)) {
 444                                 notify_writer = 0;
 445                                 continue;
 446                         }
 447
 448                         /*
 449                          * When the FIFO is less then half full notify any
 450                          * waiting writer.  WANTW can be checked while
 451                          * holding just the rlock.
 452                          */
 453                         notify_writer = 1;
 454                         if ((rpipe->pipe_state & PIPE_WANTW) == 0)
 455                                 continue;
 456                 }
 457
 458                 /*
 459                  * If the "write-side" was blocked we wake it up.  This code
 460                  * is reached either when the buffer is completely emptied
 461                  * or if it becomes more then half-empty.
 462                  *
 463                  * Pipe_state can only be modified if both the rlock and
 464                  * wlock are held.
 465                  */
 466                 if (rpipe->pipe_state & PIPE_WANTW) {
 467                         lwkt_gettoken(&rpipe->pipe_wlock);
 468                         if (rpipe->pipe_state & PIPE_WANTW) {
 469                                 rpipe->pipe_state &= ~PIPE_WANTW;
 470                                 lwkt_reltoken(&rpipe->pipe_wlock);
 471                                 wakeup(rpipe);
 472                         } else {
 473                                 lwkt_reltoken(&rpipe->pipe_wlock);
 474                         }
 475                 }
 476
 477                 /*
 478                  * Pick up our copy loop again if the writer sent data to
 479                  * us while we were messing around.
 480                  *
 481                  * On a SMP box poll up to pipe_delay nanoseconds for new
 482                  * data.  Typically a value of 2000 to 4000 is sufficient
 483                  * to eradicate most IPIs/tsleeps/wakeups when a pipe
 484                  * is used for synchronous communications with small packets,
 485                  * and 8000 or so (8uS) will pipeline large buffer xfers
 486                  * between cpus over a pipe.
 487                  *
 488                  * For synchronous communications a hit means doing a
 489                  * full Awrite-Bread-Bwrite-Aread cycle in less then 2uS,
 490                  * where as miss requiring a tsleep/wakeup sequence
 491                  * will take 7uS or more.
 492                  */
 493                 if (rpipe->pipe_buffer.windex != rpipe->pipe_buffer.rindex)
 494                         continue;
 495
 496 #if defined(SMP) && defined(_RDTSC_SUPPORTED_)
 497                 if (pipe_delay) {
 498                         int64_t tsc_target;
 499                         int good = 0;
 500
 501                         tsc_target = tsc_get_target(pipe_delay);
 502                         while (tsc_test_target(tsc_target) == 0) {
 503                                 if (rpipe->pipe_buffer.windex !=
 504                                     rpipe->pipe_buffer.rindex) {
 505                                         good = 1;
 506                                         break;
 507                                 }
 508                         }
 509                         if (good)
 510                                 continue;
 511                 }
 512 #endif
 513
 514                 /*
 515                  * Detect EOF condition, do not set error.
 516                  */
 517                 if (rpipe->pipe_state & PIPE_REOF)
 518                         break;
 519
 520                 /*
 521                  * Break if some data was read, or if this was a non-blocking
 522                  * read.
 523                  */
 524                 if (nread > 0)
 525                         break;
 526
 527                 if (nbio) {
 528                         error = EAGAIN;
 529                         break;
 530                 }
 531
 532                 /*
 533                  * Last chance, interlock with WANTR.
 534                  */
 535                 lwkt_gettoken(&rpipe->pipe_wlock);
 536                 size = rpipe->pipe_buffer.windex - rpipe->pipe_buffer.rindex;
 537                 if (size) {
 538                         lwkt_reltoken(&rpipe->pipe_wlock);
 539                         continue;
 540                 }
 541
 542                 /*
 543                  * Retest EOF - acquiring a new token can temporarily release
 544                  * tokens already held.
 545                  */
 546                 if (rpipe->pipe_state & PIPE_REOF) {
 547                         lwkt_reltoken(&rpipe->pipe_wlock);
 548                         break;
 549                 }
 550
 551                 /*
 552                  * If there is no more to read in the pipe, reset its
 553                  * pointers to the beginning.  This improves cache hit
 554                  * stats.
 555                  *
 556                  * We need both locks to modify both pointers, and there
 557                  * must also not be a write in progress or the uiomove()
 558                  * in the write might block and temporarily release
 559                  * its wlock, then reacquire and update windex.  We are
 560                  * only serialized against reads, not writes.
 561                  *
 562                  * XXX should we even bother resetting the indices?  It
 563                  *     might actually be more cache efficient not to.
 564                  */
 565                 if (rpipe->pipe_buffer.rindex == rpipe->pipe_buffer.windex &&
 566                     rpipe->pipe_wip == 0) {
 567                         rpipe->pipe_buffer.rindex = 0;
 568                         rpipe->pipe_buffer.windex = 0;
 569                 }
 570
 571                 /*
 572                  * Wait for more data.
 573                  *
 574                  * Pipe_state can only be set if both the rlock and wlock
 575                  * are held.
 576                  */
 577                 rpipe->pipe_state |= PIPE_WANTR;
 578                 tsleep_interlock(rpipe, PCATCH);
 579                 lwkt_reltoken(&rpipe->pipe_wlock);
 580                 error = tsleep(rpipe, PCATCH | PINTERLOCKED, "piperd", 0);
 581                 ++pipe_rblocked_count;
 582                 if (error)
 583                         break;
 584         }
 585         pipe_end_uio(rpipe, &rpipe->pipe_rip);
 586
 587         /*
 588          * Uptime last access time
 589          */
 590         if (error == 0 && nread)
 591                 vfs_timestamp(&rpipe->pipe_atime);
 592
 593         /*
 594          * If we drained the FIFO more then half way then handle
 595          * write blocking hysteresis.
 596          *
 597          * Note that PIPE_WANTW cannot be set by the writer without
 598          * it holding both rlock and wlock, so we can test it
 599          * while holding just rlock.
 600          */
 601         if (notify_writer) {
 602                 /*
 603                  * Synchronous blocking is done on the pipe involved
 604                  */
 605                 if (rpipe->pipe_state & PIPE_WANTW) {
 606                         lwkt_gettoken(&rpipe->pipe_wlock);
 607                         if (rpipe->pipe_state & PIPE_WANTW) {
 608                                 rpipe->pipe_state &= ~PIPE_WANTW;
 609                                 lwkt_reltoken(&rpipe->pipe_wlock);
 610                                 wakeup(rpipe);
 611                         } else {
 612                                 lwkt_reltoken(&rpipe->pipe_wlock);
 613                         }
 614                 }
 615
 616                 /*
 617                  * But we may also have to deal with a kqueue which is
 618                  * stored on the same pipe as its descriptor, so a
 619                  * EVFILT_WRITE event waiting for our side to drain will
 620                  * be on the other side.
 621                  */
 622                 lwkt_gettoken(&wpipe->pipe_wlock);
 623                 pipewakeup(wpipe, 0);
 624                 lwkt_reltoken(&wpipe->pipe_wlock);
 625         }
 626         /*size = rpipe->pipe_buffer.windex - rpipe->pipe_buffer.rindex;*/
 627         lwkt_reltoken(&rpipe->pipe_rlock);
 628
 629         return (error);
 630 }
 631
 632 static int
 633 pipe_write(struct file *fp, struct uio *uio, struct ucred *cred, int fflags)
 634 {
 635         int error;
 636         int orig_resid;
 637         int nbio;
 638         struct pipe *wpipe;
 639         struct pipe *rpipe;
 640         u_int windex;
 641         u_int space;
 642         u_int wcount;
 643         int bigwrite;
 644         int bigcount;
 645
 646         /*
 647          * Writes go to the peer.  The peer will always exist.
 648          */
 649         rpipe = (struct pipe *) fp->f_data;
 650         wpipe = rpipe->pipe_peer;
 651         lwkt_gettoken(&wpipe->pipe_wlock);
 652         if (wpipe->pipe_state & PIPE_WEOF) {
 653                 lwkt_reltoken(&wpipe->pipe_wlock);
 654                 return (EPIPE);
 655         }
 656
 657         /*
 658          * Degenerate case (EPIPE takes prec)
 659          */
 660         if (uio->uio_resid == 0) {
 661                 lwkt_reltoken(&wpipe->pipe_wlock);
 662                 return(0);
 663         }
 664
 665         /*
 666          * Writes are serialized (start_uio must be called with wlock)
 667          */
 668         error = pipe_start_uio(wpipe, &wpipe->pipe_wip);
 669         if (error) {
 670                 lwkt_reltoken(&wpipe->pipe_wlock);
 671                 return (error);
 672         }
 673
 674         if (fflags & O_FBLOCKING)
 675                 nbio = 0;
 676         else if (fflags & O_FNONBLOCKING)
 677                 nbio = 1;
 678         else if (fp->f_flag & O_NONBLOCK)
 679                 nbio = 1;
 680         else
 681                 nbio = 0;
 682
 683         /*
 684          * If it is advantageous to resize the pipe buffer, do
 685          * so.  We are write-serialized so we can block safely.
 686          */
 687         if ((wpipe->pipe_buffer.size <= PIPE_SIZE) &&
 688             (pipe_nbig < pipe_maxbig) &&
 689             wpipe->pipe_wantwcnt > 4 &&
 690             (wpipe->pipe_buffer.rindex == wpipe->pipe_buffer.windex)) {
 691                 /*
 692                  * Recheck after lock.
 693                  */
 694                 lwkt_gettoken(&wpipe->pipe_rlock);
 695                 if ((wpipe->pipe_buffer.size <= PIPE_SIZE) &&
 696                     (pipe_nbig < pipe_maxbig) &&
 697                     (wpipe->pipe_buffer.rindex == wpipe->pipe_buffer.windex)) {
 698                         atomic_add_int(&pipe_nbig, 1);
 699                         if (pipespace(wpipe, BIG_PIPE_SIZE) == 0)
 700                                 ++pipe_bigcount;
 701                         else
 702                                 atomic_subtract_int(&pipe_nbig, 1);
 703                 }
 704                 lwkt_reltoken(&wpipe->pipe_rlock);
 705         }
 706
 707         orig_resid = uio->uio_resid;
 708         wcount = 0;
 709
 710         bigwrite = (uio->uio_resid > 10 * 1024 * 1024);
 711         bigcount = 10;
 712
 713         while (uio->uio_resid) {
 714                 if (wpipe->pipe_state & PIPE_WEOF) {
 715                         error = EPIPE;
 716                         break;
 717                 }
 718
 719                 /*
 720                  * Don't hog the cpu.
 721                  */
 722                 if (bigwrite && --bigcount == 0) {
 723                         lwkt_user_yield();
 724                         bigcount = 10;
 725                         if (CURSIG(curthread->td_lwp)) {
 726                                 error = EINTR;
 727                                 break;
 728                         }
 729                 }
 730
 731                 windex = wpipe->pipe_buffer.windex &
 732                          (wpipe->pipe_buffer.size - 1);
 733                 space = wpipe->pipe_buffer.size -
 734                         (wpipe->pipe_buffer.windex - wpipe->pipe_buffer.rindex);
 735                 cpu_lfence();
 736
 737                 /* Writes of size <= PIPE_BUF must be atomic. */
 738                 if ((space < uio->uio_resid) && (orig_resid <= PIPE_BUF))
 739                         space = 0;
 740
 741                 /*
 742                  * Write to fill, read size handles write hysteresis.  Also
 743                  * additional restrictions can cause select-based non-blocking
 744                  * writes to spin.
 745                  */
 746                 if (space > 0) {
 747                         u_int segsize;
 748
 749                         /*
 750                          * Transfer size is minimum of uio transfer
 751                          * and free space in pipe buffer.
 752                          *
 753                          * Limit each uiocopy to no more then PIPE_SIZE
 754                          * so we can keep the gravy train going on a
 755                          * SMP box.  This doubles the performance for
 756                          * write sizes > 16K.  Otherwise large writes
 757                          * wind up doing an inefficient synchronous
 758                          * ping-pong.
 759                          */
 760                         space = szmin(space, uio->uio_resid);
 761                         if (space > PIPE_SIZE)
 762                                 space = PIPE_SIZE;
 763
 764                         /*
 765                          * First segment to transfer is minimum of
 766                          * transfer size and contiguous space in
 767                          * pipe buffer.  If first segment to transfer
 768                          * is less than the transfer size, we've got
 769                          * a wraparound in the buffer.
 770                          */
 771                         segsize = wpipe->pipe_buffer.size - windex;
 772                         if (segsize > space)
 773                                 segsize = space;
 774
 775 #ifdef SMP
 776                         /*
 777                          * If this is the first loop and the reader is
 778                          * blocked, do a preemptive wakeup of the reader.
 779                          *
 780                          * On SMP the IPI latency plus the wlock interlock
 781                          * on the reader side is the fastest way to get the
 782                          * reader going.  (The scheduler will hard loop on
 783                          * lock tokens).
 784                          *
 785                          * NOTE: We can't clear WANTR here without acquiring
 786                          * the rlock, which we don't want to do here!
 787                          */
 788                         if ((wpipe->pipe_state & PIPE_WANTR))
 789                                 wakeup(wpipe);
 790 #endif
 791
 792                         /*
 793                          * Transfer segment, which may include a wrap-around.
 794                          * Update windex to account for both all in one go
 795                          * so the reader can read() the data atomically.
 796                          */
 797                         error = uiomove(&wpipe->pipe_buffer.buffer[windex],
 798                                         segsize, uio);
 799                         if (error == 0 && segsize < space) {
 800                                 segsize = space - segsize;
 801                                 error = uiomove(&wpipe->pipe_buffer.buffer[0],
 802                                                 segsize, uio);
 803                         }
 804                         if (error)
 805                                 break;
 806                         cpu_mfence();
 807                         wpipe->pipe_buffer.windex += space;
 808                         wcount += space;
 809                         continue;
 810                 }
 811
 812                 /*
 813                  * We need both the rlock and the wlock to interlock against
 814                  * the EOF, WANTW, and size checks, and to modify pipe_state.
 815                  *
 816                  * These are token locks so we do not have to worry about
 817                  * deadlocks.
 818                  */
 819                 lwkt_gettoken(&wpipe->pipe_rlock);
 820
 821                 /*
 822                  * If the "read-side" has been blocked, wake it up now
 823                  * and yield to let it drain synchronously rather
 824                  * then block.
 825                  */
 826                 if (wpipe->pipe_state & PIPE_WANTR) {
 827                         wpipe->pipe_state &= ~PIPE_WANTR;
 828                         wakeup(wpipe);
 829                 }
 830
 831                 /*
 832                  * don't block on non-blocking I/O
 833                  */
 834                 if (nbio) {
 835                         lwkt_reltoken(&wpipe->pipe_rlock);
 836                         error = EAGAIN;
 837                         break;
 838                 }
 839
 840                 /*
 841                  * re-test whether we have to block in the writer after
 842                  * acquiring both locks, in case the reader opened up
 843                  * some space.
 844                  */
 845                 space = wpipe->pipe_buffer.size -
 846                         (wpipe->pipe_buffer.windex - wpipe->pipe_buffer.rindex);
 847                 cpu_lfence();
 848                 if ((space < uio->uio_resid) && (orig_resid <= PIPE_BUF))
 849                         space = 0;
 850
 851                 /*
 852                  * Retest EOF - acquiring a new token can temporarily release
 853                  * tokens already held.
 854                  */
 855                 if (wpipe->pipe_state & PIPE_WEOF) {
 856                         lwkt_reltoken(&wpipe->pipe_rlock);
 857                         error = EPIPE;
 858                         break;
 859                 }
 860
 861                 /*
 862                  * We have no more space and have something to offer,
 863                  * wake up select/poll/kq.
 864                  */
 865                 if (space == 0) {
 866                         wpipe->pipe_state |= PIPE_WANTW;
 867                         ++wpipe->pipe_wantwcnt;
 868                         pipewakeup(wpipe, 1);
 869                         if (wpipe->pipe_state & PIPE_WANTW)
 870                                 error = tsleep(wpipe, PCATCH, "pipewr", 0);
 871                         ++pipe_wblocked_count;
 872                 }
 873                 lwkt_reltoken(&wpipe->pipe_rlock);
 874
 875                 /*
 876                  * Break out if we errored or the read side wants us to go
 877                  * away.
 878                  */
 879                 if (error)
 880                         break;
 881                 if (wpipe->pipe_state & PIPE_WEOF) {
 882                         error = EPIPE;
 883                         break;
 884                 }
 885         }
 886         pipe_end_uio(wpipe, &wpipe->pipe_wip);
 887
 888         /*
 889          * If we have put any characters in the buffer, we wake up
 890          * the reader.
 891          *
 892          * Both rlock and wlock are required to be able to modify pipe_state.
 893          */
 894         if (wpipe->pipe_buffer.windex != wpipe->pipe_buffer.rindex) {
 895                 if (wpipe->pipe_state & PIPE_WANTR) {
 896                         lwkt_gettoken(&wpipe->pipe_rlock);
 897                         if (wpipe->pipe_state & PIPE_WANTR) {
 898                                 wpipe->pipe_state &= ~PIPE_WANTR;
 899                                 lwkt_reltoken(&wpipe->pipe_rlock);
 900                                 wakeup(wpipe);
 901                         } else {
 902                                 lwkt_reltoken(&wpipe->pipe_rlock);
 903                         }
 904                 }
 905                 lwkt_gettoken(&wpipe->pipe_rlock);
 906                 pipewakeup(wpipe, 1);
 907                 lwkt_reltoken(&wpipe->pipe_rlock);
 908         }
 909
 910         /*
 911          * Don't return EPIPE if I/O was successful
 912          */
 913         if ((wpipe->pipe_buffer.rindex == wpipe->pipe_buffer.windex) &&
 914             (uio->uio_resid == 0) &&
 915             (error == EPIPE)) {
 916                 error = 0;
 917         }
 918
 919         if (error == 0)
 920                 vfs_timestamp(&wpipe->pipe_mtime);
 921
 922         /*
 923          * We have something to offer,
 924          * wake up select/poll/kq.
 925          */
 926         /*space = wpipe->pipe_buffer.windex - wpipe->pipe_buffer.rindex;*/
 927         lwkt_reltoken(&wpipe->pipe_wlock);
 928         return (error);
 929 }
 930
 931 /*
 932  * we implement a very minimal set of ioctls for compatibility with sockets.
 933  */
 934 int
 935 pipe_ioctl(struct file *fp, u_long cmd, caddr_t data,
 936            struct ucred *cred, struct sysmsg *msg)
 937 {
 938         struct pipe *mpipe;
 939         int error;
 940
 941         mpipe = (struct pipe *)fp->f_data;
 942
 943         lwkt_gettoken(&mpipe->pipe_rlock);
 944         lwkt_gettoken(&mpipe->pipe_wlock);
 945
 946         switch (cmd) {
 947         case FIOASYNC:
 948                 if (*(int *)data) {
 949                         mpipe->pipe_state |= PIPE_ASYNC;
 950                 } else {
 951                         mpipe->pipe_state &= ~PIPE_ASYNC;
 952                 }
 953                 error = 0;
 954                 break;
 955         case FIONREAD:
 956                 *(int *)data = mpipe->pipe_buffer.windex -
 957                                 mpipe->pipe_buffer.rindex;
 958                 error = 0;
 959                 break;
 960         case FIOSETOWN:
 961                 error = fsetown(*(int *)data, &mpipe->pipe_sigio);
 962                 break;
 963         case FIOGETOWN:
 964                 *(int *)data = fgetown(&mpipe->pipe_sigio);
 965                 error = 0;
 966                 break;
 967         case TIOCSPGRP:
 968                 /* This is deprecated, FIOSETOWN should be used instead. */
 969                 error = fsetown(-(*(int *)data), &mpipe->pipe_sigio);
 970                 break;
 971
 972         case TIOCGPGRP:
 973                 /* This is deprecated, FIOGETOWN should be used instead. */
 974                 *(int *)data = -fgetown(&mpipe->pipe_sigio);
 975                 error = 0;
 976                 break;
 977         default:
 978                 error = ENOTTY;
 979                 break;
 980         }
 981         lwkt_reltoken(&mpipe->pipe_wlock);
 982         lwkt_reltoken(&mpipe->pipe_rlock);
 983
 984         return (error);
 985 }
 986
 987 /*
 988  * MPSAFE
 989  */
 990 static int
 991 pipe_stat(struct file *fp, struct stat *ub, struct ucred *cred)
 992 {
 993         struct pipe *pipe;
 994
 995         pipe = (struct pipe *)fp->f_data;
 996
 997         bzero((caddr_t)ub, sizeof(*ub));
 998         ub->st_mode = S_IFIFO;
 999         ub->st_blksize = pipe->pipe_buffer.size;
1000         ub->st_size = pipe->pipe_buffer.windex - pipe->pipe_buffer.rindex;
1001         ub->st_blocks = (ub->st_size + ub->st_blksize - 1) / ub->st_blksize;
1002         ub->st_atimespec = pipe->pipe_atime;
1003         ub->st_mtimespec = pipe->pipe_mtime;
1004         ub->st_ctimespec = pipe->pipe_ctime;
1005         /*
1006          * Left as 0: st_dev, st_ino, st_nlink, st_uid, st_gid, st_rdev,
1007          * st_flags, st_gen.
1008          * XXX (st_dev, st_ino) should be unique.
1009          */
1010         return (0);
1011 }
1012
1013 static int
1014 pipe_close(struct file *fp)
1015 {
1016         struct pipe *cpipe;
1017
1018         cpipe = (struct pipe *)fp->f_data;
1019         fp->f_ops = &badfileops;
1020         fp->f_data = NULL;
1021         funsetown(&cpipe->pipe_sigio);
1022         pipeclose(cpipe);
1023         return (0);
1024 }
1025
1026 /*
1027  * Shutdown one or both directions of a full-duplex pipe.
1028  */
1029 static int
1030 pipe_shutdown(struct file *fp, int how)
1031 {
1032         struct pipe *rpipe;
1033         struct pipe *wpipe;
1034         int error = EPIPE;
1035
1036         rpipe = (struct pipe *)fp->f_data;
1037         wpipe = rpipe->pipe_peer;
1038
1039         /*
1040          * We modify pipe_state on both pipes, which means we need
1041          * all four tokens!
1042          */
1043         lwkt_gettoken(&rpipe->pipe_rlock);
1044         lwkt_gettoken(&rpipe->pipe_wlock);
1045         lwkt_gettoken(&wpipe->pipe_rlock);
1046         lwkt_gettoken(&wpipe->pipe_wlock);
1047
1048         switch(how) {
1049         case SHUT_RDWR:
1050         case SHUT_RD:
1051                 rpipe->pipe_state |= PIPE_REOF;         /* my reads */
1052                 rpipe->pipe_state |= PIPE_WEOF;         /* peer writes */
1053                 if (rpipe->pipe_state & PIPE_WANTR) {
1054                         rpipe->pipe_state &= ~PIPE_WANTR;
1055                         wakeup(rpipe);
1056                 }
1057                 if (rpipe->pipe_state & PIPE_WANTW) {
1058                         rpipe->pipe_state &= ~PIPE_WANTW;
1059                         wakeup(rpipe);
1060                 }
1061                 error = 0;
1062                 if (how == SHUT_RD)
1063                         break;
1064                 /* fall through */
1065         case SHUT_WR:
1066                 wpipe->pipe_state |= PIPE_REOF;         /* peer reads */
1067                 wpipe->pipe_state |= PIPE_WEOF;         /* my writes */
1068                 if (wpipe->pipe_state & PIPE_WANTR) {
1069                         wpipe->pipe_state &= ~PIPE_WANTR;
1070                         wakeup(wpipe);
1071                 }
1072                 if (wpipe->pipe_state & PIPE_WANTW) {
1073                         wpipe->pipe_state &= ~PIPE_WANTW;
1074                         wakeup(wpipe);
1075                 }
1076                 error = 0;
1077                 break;
1078         }
1079         pipewakeup(rpipe, 1);
1080         pipewakeup(wpipe, 1);
1081
1082         lwkt_reltoken(&wpipe->pipe_wlock);
1083         lwkt_reltoken(&wpipe->pipe_rlock);
1084         lwkt_reltoken(&rpipe->pipe_wlock);
1085         lwkt_reltoken(&rpipe->pipe_rlock);
1086
1087         return (error);
1088 }
1089
1090 static void
1091 pipe_free_kmem(struct pipe *cpipe)
1092 {
1093         if (cpipe->pipe_buffer.buffer != NULL) {
1094                 if (cpipe->pipe_buffer.size > PIPE_SIZE)
1095                         atomic_subtract_int(&pipe_nbig, 1);
1096                 kmem_free(&kernel_map,
1097                         (vm_offset_t)cpipe->pipe_buffer.buffer,
1098                         cpipe->pipe_buffer.size);
1099                 cpipe->pipe_buffer.buffer = NULL;
1100                 cpipe->pipe_buffer.object = NULL;
1101         }
1102 }
1103
1104 /*
1105  * Close the pipe.  The slock must be held to interlock against simultanious
1106  * closes.  The rlock and wlock must be held to adjust the pipe_state.
1107  */
1108 static void
1109 pipeclose(struct pipe *cpipe)
1110 {
1111         globaldata_t gd;
1112         struct pipe *ppipe;
1113
1114         if (cpipe == NULL)
1115                 return;
1116
1117         /*
1118          * The slock may not have been allocated yet (close during
1119          * initialization)
1120          *
1121          * We need both the read and write tokens to modify pipe_state.
1122          */
1123         if (cpipe->pipe_slock)
1124                 lockmgr(cpipe->pipe_slock, LK_EXCLUSIVE);
1125         lwkt_gettoken(&cpipe->pipe_rlock);
1126         lwkt_gettoken(&cpipe->pipe_wlock);
1127
1128         /*
1129          * Set our state, wakeup anyone waiting in select/poll/kq, and
1130          * wakeup anyone blocked on our pipe.
1131          */
1132         cpipe->pipe_state |= PIPE_CLOSED | PIPE_REOF | PIPE_WEOF;
1133         pipewakeup(cpipe, 1);
1134         if (cpipe->pipe_state & (PIPE_WANTR | PIPE_WANTW)) {
1135                 cpipe->pipe_state &= ~(PIPE_WANTR | PIPE_WANTW);
1136                 wakeup(cpipe);
1137         }
1138
1139         /*
1140          * Disconnect from peer.
1141          */
1142         if ((ppipe = cpipe->pipe_peer) != NULL) {
1143                 lwkt_gettoken(&ppipe->pipe_rlock);
1144                 lwkt_gettoken(&ppipe->pipe_wlock);
1145                 ppipe->pipe_state |= PIPE_REOF | PIPE_WEOF;
1146                 pipewakeup(ppipe, 1);
1147                 if (ppipe->pipe_state & (PIPE_WANTR | PIPE_WANTW)) {
1148                         ppipe->pipe_state &= ~(PIPE_WANTR | PIPE_WANTW);
1149                         wakeup(ppipe);
1150                 }
1151                 if (SLIST_FIRST(&ppipe->pipe_kq.ki_note))
1152                         KNOTE(&ppipe->pipe_kq.ki_note, 0);
1153                 lwkt_reltoken(&ppipe->pipe_wlock);
1154                 lwkt_reltoken(&ppipe->pipe_rlock);
1155         }
1156
1157         /*
1158          * If the peer is also closed we can free resources for both
1159          * sides, otherwise we leave our side intact to deal with any
1160          * races (since we only have the slock).
1161          */
1162         if (ppipe && (ppipe->pipe_state & PIPE_CLOSED)) {
1163                 cpipe->pipe_peer = NULL;
1164                 ppipe->pipe_peer = NULL;
1165                 ppipe->pipe_slock = NULL;       /* we will free the slock */
1166                 pipeclose(ppipe);
1167                 ppipe = NULL;
1168         }
1169
1170         lwkt_reltoken(&cpipe->pipe_wlock);
1171         lwkt_reltoken(&cpipe->pipe_rlock);
1172         if (cpipe->pipe_slock)
1173                 lockmgr(cpipe->pipe_slock, LK_RELEASE);
1174
1175         /*
1176          * If we disassociated from our peer we can free resources
1177          */
1178         if (ppipe == NULL) {
1179                 gd = mycpu;
1180                 if (cpipe->pipe_slock) {
1181                         kfree(cpipe->pipe_slock, M_PIPE);
1182                         cpipe->pipe_slock = NULL;
1183                 }
1184                 if (gd->gd_pipeqcount >= pipe_maxcache ||
1185                     cpipe->pipe_buffer.size != PIPE_SIZE
1186                 ) {
1187                         pipe_free_kmem(cpipe);
1188                         kfree(cpipe, M_PIPE);
1189                 } else {
1190                         cpipe->pipe_state = 0;
1191                         cpipe->pipe_peer = gd->gd_pipeq;
1192                         gd->gd_pipeq = cpipe;
1193                         ++gd->gd_pipeqcount;
1194                 }
1195         }
1196 }
1197
1198 static int
1199 pipe_kqfilter(struct file *fp, struct knote *kn)
1200 {
1201         struct pipe *cpipe;
1202
1203         cpipe = (struct pipe *)kn->kn_fp->f_data;
1204
1205         switch (kn->kn_filter) {
1206         case EVFILT_READ:
1207                 kn->kn_fop = &pipe_rfiltops;
1208                 break;
1209         case EVFILT_WRITE:
1210                 kn->kn_fop = &pipe_wfiltops;
1211                 if (cpipe->pipe_peer == NULL) {
1212                         /* other end of pipe has been closed */
1213                         return (EPIPE);
1214                 }
1215                 break;
1216         default:
1217                 return (EOPNOTSUPP);
1218         }
1219         kn->kn_hook = (caddr_t)cpipe;
1220
1221         knote_insert(&cpipe->pipe_kq.ki_note, kn);
1222
1223         return (0);
1224 }
1225
1226 static void
1227 filt_pipedetach(struct knote *kn)
1228 {
1229         struct pipe *cpipe = (struct pipe *)kn->kn_hook;
1230
1231         knote_remove(&cpipe->pipe_kq.ki_note, kn);
1232 }
1233
1234 /*ARGSUSED*/
1235 static int
1236 filt_piperead(struct knote *kn, long hint)
1237 {
1238         struct pipe *rpipe = (struct pipe *)kn->kn_fp->f_data;
1239         int ready = 0;
1240
1241         lwkt_gettoken(&rpipe->pipe_rlock);
1242         lwkt_gettoken(&rpipe->pipe_wlock);
1243
1244         kn->kn_data = rpipe->pipe_buffer.windex - rpipe->pipe_buffer.rindex;
1245
1246         /*
1247          * Only set EOF if all data has been exhausted
1248          */
1249         if ((rpipe->pipe_state & PIPE_REOF) && kn->kn_data == 0) {
1250                 kn->kn_flags |= EV_EOF;
1251                 ready = 1;
1252         }
1253
1254         lwkt_reltoken(&rpipe->pipe_wlock);
1255         lwkt_reltoken(&rpipe->pipe_rlock);
1256
1257         if (!ready)
1258                 ready = kn->kn_data > 0;
1259
1260         return (ready);
1261 }
1262
1263 /*ARGSUSED*/
1264 static int
1265 filt_pipewrite(struct knote *kn, long hint)
1266 {
1267         struct pipe *rpipe = (struct pipe *)kn->kn_fp->f_data;
1268         struct pipe *wpipe = rpipe->pipe_peer;
1269         int ready = 0;
1270
1271         kn->kn_data = 0;
1272         if (wpipe == NULL) {
1273                 kn->kn_flags |= EV_EOF;
1274                 return (1);
1275         }
1276
1277         lwkt_gettoken(&wpipe->pipe_rlock);
1278         lwkt_gettoken(&wpipe->pipe_wlock);
1279
1280         if (wpipe->pipe_state & PIPE_WEOF) {
1281                 kn->kn_flags |= EV_EOF;
1282                 ready = 1;
1283         }
1284
1285         if (!ready)
1286                 kn->kn_data = wpipe->pipe_buffer.size -
1287                               (wpipe->pipe_buffer.windex -
1288                                wpipe->pipe_buffer.rindex);
1289
1290         lwkt_reltoken(&wpipe->pipe_wlock);
1291         lwkt_reltoken(&wpipe->pipe_rlock);
1292
1293         if (!ready)
1294                 ready = kn->kn_data >= PIPE_BUF;
1295
1296         return (ready);
1297 }