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  */
  21
  22 /*
  23  * This file contains a high-performance replacement for the socket-based
  24  * pipes scheme originally used in FreeBSD/4.4Lite.  It does not support
  25  * all features of sockets, but does do everything that pipes normally
  26  * do.
  27  */
  28 #include <sys/param.h>
  29 #include <sys/systm.h>
  30 #include <sys/kernel.h>
  31 #include <sys/proc.h>
  32 #include <sys/fcntl.h>
  33 #include <sys/file.h>
  34 #include <sys/filedesc.h>
  35 #include <sys/filio.h>
  36 #include <sys/ttycom.h>
  37 #include <sys/stat.h>
  38 #include <sys/signalvar.h>
  39 #include <sys/sysproto.h>
  40 #include <sys/pipe.h>
  41 #include <sys/vnode.h>
  42 #include <sys/uio.h>
  43 #include <sys/event.h>
  44 #include <sys/globaldata.h>
  45 #include <sys/module.h>
  46 #include <sys/malloc.h>
  47 #include <sys/sysctl.h>
  48 #include <sys/socket.h>
  49 #include <sys/kern_syscall.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, "number 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 static int pipe_delay = 5000;   /* 5uS default */
 140 SYSCTL_INT(_kern_pipe, OID_AUTO, delay,
 141         CTLFLAG_RW, &pipe_delay, 0, "SMP delay optimization in ns");
 142 #if !defined(NO_PIPE_SYSCTL_STATS)
 143 SYSCTL_INT(_kern_pipe, OID_AUTO, bcache_alloc,
 144         CTLFLAG_RW, &pipe_bcache_alloc, 0, "pipe buffer from pcpu cache");
 145 SYSCTL_INT(_kern_pipe, OID_AUTO, bkmem_alloc,
 146         CTLFLAG_RW, &pipe_bkmem_alloc, 0, "pipe buffer from kmem");
 147 #endif
 148
 149 /*
 150  * Auto-size pipe cache to reduce kmem allocations and frees.
 151  */
 152 static
 153 void
 154 pipeinit(void *dummy)
 155 {
 156         size_t mbytes = kmem_lim_size();
 157
 158         if (pipe_maxbig == LIMITBIGPIPES) {
 159                 if (mbytes >= 7 * 1024)
 160                         pipe_maxbig *= 2;
 161                 if (mbytes >= 15 * 1024)
 162                         pipe_maxbig *= 2;
 163         }
 164         if (pipe_maxcache == PIPEQ_MAX_CACHE) {
 165                 if (mbytes >= 7 * 1024)
 166                         pipe_maxcache *= 2;
 167                 if (mbytes >= 15 * 1024)
 168                         pipe_maxcache *= 2;
 169         }
 170 }
 171 SYSINIT(kmem, SI_BOOT2_MACHDEP, SI_ORDER_ANY, pipeinit, NULL);
 172
 173 static void pipeclose (struct pipe *cpipe);
 174 static void pipe_free_kmem (struct pipe *cpipe);
 175 static int pipe_create (struct pipe **cpipep);
 176 static int pipespace (struct pipe *cpipe, int size);
 177
 178 static __inline void
 179 pipewakeup(struct pipe *cpipe, int dosigio)
 180 {
 181         if (dosigio && (cpipe->pipe_state & PIPE_ASYNC) && cpipe->pipe_sigio) {
 182                 lwkt_gettoken(&sigio_token);
 183                 pgsigio(cpipe->pipe_sigio, SIGIO, 0);
 184                 lwkt_reltoken(&sigio_token);
 185         }
 186         KNOTE(&cpipe->pipe_kq.ki_note, 0);
 187 }
 188
 189 /*
 190  * These routines are called before and after a UIO.  The UIO
 191  * may block, causing our held tokens to be lost temporarily.
 192  *
 193  * We use these routines to serialize reads against other reads
 194  * and writes against other writes.
 195  *
 196  * The read token is held on entry so *ipp does not race.
 197  */
 198 static __inline int
 199 pipe_start_uio(struct pipe *cpipe, int *ipp)
 200 {
 201         int error;
 202
 203         while (*ipp) {
 204                 *ipp = -1;
 205                 error = tsleep(ipp, PCATCH, "pipexx", 0);
 206                 if (error)
 207                         return (error);
 208         }
 209         *ipp = 1;
 210         return (0);
 211 }
 212
 213 static __inline void
 214 pipe_end_uio(struct pipe *cpipe, int *ipp)
 215 {
 216         if (*ipp < 0) {
 217                 *ipp = 0;
 218                 wakeup(ipp);
 219         } else {
 220                 KKASSERT(*ipp > 0);
 221                 *ipp = 0;
 222         }
 223 }
 224
 225 /*
 226  * The pipe system call for the DTYPE_PIPE type of pipes
 227  *
 228  * pipe_args(int dummy)
 229  *
 230  * MPSAFE
 231  */
 232 int
 233 sys_pipe(struct pipe_args *uap)
 234 {
 235         return kern_pipe(uap->sysmsg_fds, 0);
 236 }
 237
 238 int
 239 sys_pipe2(struct pipe2_args *uap)
 240 {
 241         return kern_pipe(uap->sysmsg_fds, uap->flags);
 242 }
 243
 244 int
 245 kern_pipe(long *fds, int flags)
 246 {
 247         struct thread *td = curthread;
 248         struct filedesc *fdp = td->td_proc->p_fd;
 249         struct file *rf, *wf;
 250         struct pipe *rpipe, *wpipe;
 251         int fd1, fd2, error;
 252
 253         rpipe = wpipe = NULL;
 254         if (pipe_create(&rpipe) || pipe_create(&wpipe)) {
 255                 pipeclose(rpipe);
 256                 pipeclose(wpipe);
 257                 return (ENFILE);
 258         }
 259
 260         error = falloc(td->td_lwp, &rf, &fd1);
 261         if (error) {
 262                 pipeclose(rpipe);
 263                 pipeclose(wpipe);
 264                 return (error);
 265         }
 266         fds[0] = fd1;
 267
 268         /*
 269          * Warning: once we've gotten past allocation of the fd for the
 270          * read-side, we can only drop the read side via fdrop() in order
 271          * to avoid races against processes which manage to dup() the read
 272          * side while we are blocked trying to allocate the write side.
 273          */
 274         rf->f_type = DTYPE_PIPE;
 275         rf->f_flag = FREAD | FWRITE;
 276         rf->f_ops = &pipeops;
 277         rf->f_data = rpipe;
 278         if (flags & O_NONBLOCK)
 279                 rf->f_flag |= O_NONBLOCK;
 280         if (flags & O_CLOEXEC)
 281                 fdp->fd_files[fd1].fileflags |= UF_EXCLOSE;
 282
 283         error = falloc(td->td_lwp, &wf, &fd2);
 284         if (error) {
 285                 fsetfd(fdp, NULL, fd1);
 286                 fdrop(rf);
 287                 /* rpipe has been closed by fdrop(). */
 288                 pipeclose(wpipe);
 289                 return (error);
 290         }
 291         wf->f_type = DTYPE_PIPE;
 292         wf->f_flag = FREAD | FWRITE;
 293         wf->f_ops = &pipeops;
 294         wf->f_data = wpipe;
 295         if (flags & O_NONBLOCK)
 296                 wf->f_flag |= O_NONBLOCK;
 297         if (flags & O_CLOEXEC)
 298                 fdp->fd_files[fd2].fileflags |= UF_EXCLOSE;
 299
 300         fds[1] = fd2;
 301
 302         rpipe->pipe_slock = kmalloc(sizeof(struct lock),
 303                                     M_PIPE, M_WAITOK|M_ZERO);
 304         wpipe->pipe_slock = rpipe->pipe_slock;
 305         rpipe->pipe_peer = wpipe;
 306         wpipe->pipe_peer = rpipe;
 307         lockinit(rpipe->pipe_slock, "pipecl", 0, 0);
 308
 309         /*
 310          * Once activated the peer relationship remains valid until
 311          * both sides are closed.
 312          */
 313         fsetfd(fdp, rf, fd1);
 314         fsetfd(fdp, wf, fd2);
 315         fdrop(rf);
 316         fdrop(wf);
 317
 318         return (0);
 319 }
 320
 321 /*
 322  * Allocate kva for pipe circular buffer, the space is pageable
 323  * This routine will 'realloc' the size of a pipe safely, if it fails
 324  * it will retain the old buffer.
 325  * If it fails it will return ENOMEM.
 326  */
 327 static int
 328 pipespace(struct pipe *cpipe, int size)
 329 {
 330         struct vm_object *object;
 331         caddr_t buffer;
 332         int npages, error;
 333
 334         npages = round_page(size) / PAGE_SIZE;
 335         object = cpipe->pipe_buffer.object;
 336
 337         /*
 338          * [re]create the object if necessary and reserve space for it
 339          * in the kernel_map.  The object and memory are pageable.  On
 340          * success, free the old resources before assigning the new
 341          * ones.
 342          */
 343         if (object == NULL || object->size != npages) {
 344                 object = vm_object_allocate(OBJT_DEFAULT, npages);
 345                 buffer = (caddr_t)vm_map_min(&kernel_map);
 346
 347                 error = vm_map_find(&kernel_map, object, NULL,
 348                                     0, (vm_offset_t *)&buffer, size,
 349                                     PAGE_SIZE,
 350                                     1, VM_MAPTYPE_NORMAL,
 351                                     VM_PROT_ALL, VM_PROT_ALL, 0);
 352
 353                 if (error != KERN_SUCCESS) {
 354                         vm_object_deallocate(object);
 355                         return (ENOMEM);
 356                 }
 357                 pipe_free_kmem(cpipe);
 358                 cpipe->pipe_buffer.object = object;
 359                 cpipe->pipe_buffer.buffer = buffer;
 360                 cpipe->pipe_buffer.size = size;
 361                 ++pipe_bkmem_alloc;
 362         } else {
 363                 ++pipe_bcache_alloc;
 364         }
 365         cpipe->pipe_buffer.rindex = 0;
 366         cpipe->pipe_buffer.windex = 0;
 367         return (0);
 368 }
 369
 370 /*
 371  * Initialize and allocate VM and memory for pipe, pulling the pipe from
 372  * our per-cpu cache if possible.  For now make sure it is sized for the
 373  * smaller PIPE_SIZE default.
 374  */
 375 static int
 376 pipe_create(struct pipe **cpipep)
 377 {
 378         globaldata_t gd = mycpu;
 379         struct pipe *cpipe;
 380         int error;
 381
 382         if ((cpipe = gd->gd_pipeq) != NULL) {
 383                 gd->gd_pipeq = cpipe->pipe_peer;
 384                 --gd->gd_pipeqcount;
 385                 cpipe->pipe_peer = NULL;
 386                 cpipe->pipe_wantwcnt = 0;
 387         } else {
 388                 cpipe = kmalloc(sizeof(struct pipe), M_PIPE, M_WAITOK|M_ZERO);
 389         }
 390         *cpipep = cpipe;
 391         if ((error = pipespace(cpipe, PIPE_SIZE)) != 0)
 392                 return (error);
 393         vfs_timestamp(&cpipe->pipe_ctime);
 394         cpipe->pipe_atime = cpipe->pipe_ctime;
 395         cpipe->pipe_mtime = cpipe->pipe_ctime;
 396         lwkt_token_init(&cpipe->pipe_rlock, "piper");
 397         lwkt_token_init(&cpipe->pipe_wlock, "pipew");
 398         return (0);
 399 }
 400
 401 static int
 402 pipe_read(struct file *fp, struct uio *uio, struct ucred *cred, int fflags)
 403 {
 404         struct pipe *rpipe;
 405         struct pipe *wpipe;
 406         int error;
 407         size_t nread = 0;
 408         int nbio;
 409         u_int size;     /* total bytes available */
 410         u_int nsize;    /* total bytes to read */
 411         u_int rindex;   /* contiguous bytes available */
 412         int notify_writer;
 413         int bigread;
 414         int bigcount;
 415
 416         atomic_set_int(&curthread->td_mpflags, TDF_MP_BATCH_DEMARC);
 417
 418         if (uio->uio_resid == 0)
 419                 return(0);
 420
 421         /*
 422          * Setup locks, calculate nbio
 423          */
 424         rpipe = (struct pipe *)fp->f_data;
 425         wpipe = rpipe->pipe_peer;
 426         lwkt_gettoken(&rpipe->pipe_rlock);
 427
 428         if (fflags & O_FBLOCKING)
 429                 nbio = 0;
 430         else if (fflags & O_FNONBLOCKING)
 431                 nbio = 1;
 432         else if (fp->f_flag & O_NONBLOCK)
 433                 nbio = 1;
 434         else
 435                 nbio = 0;
 436
 437         /*
 438          * Reads are serialized.  Note however that pipe_buffer.buffer and
 439          * pipe_buffer.size can change out from under us when the number
 440          * of bytes in the buffer are zero due to the write-side doing a
 441          * pipespace().
 442          */
 443         error = pipe_start_uio(rpipe, &rpipe->pipe_rip);
 444         if (error) {
 445                 lwkt_reltoken(&rpipe->pipe_rlock);
 446                 return (error);
 447         }
 448         notify_writer = 0;
 449
 450         bigread = (uio->uio_resid > 10 * 1024 * 1024);
 451         bigcount = 10;
 452
 453         while (uio->uio_resid) {
 454                 /*
 455                  * Don't hog the cpu.
 456                  */
 457                 if (bigread && --bigcount == 0) {
 458                         lwkt_user_yield();
 459                         bigcount = 10;
 460                         if (CURSIG(curthread->td_lwp)) {
 461                                 error = EINTR;
 462                                 break;
 463                         }
 464                 }
 465
 466                 size = rpipe->pipe_buffer.windex - rpipe->pipe_buffer.rindex;
 467                 cpu_lfence();
 468                 if (size) {
 469                         rindex = rpipe->pipe_buffer.rindex &
 470                                  (rpipe->pipe_buffer.size - 1);
 471                         nsize = size;
 472                         if (nsize > rpipe->pipe_buffer.size - rindex)
 473                                 nsize = rpipe->pipe_buffer.size - rindex;
 474                         nsize = szmin(nsize, uio->uio_resid);
 475
 476                         error = uiomove(&rpipe->pipe_buffer.buffer[rindex],
 477                                         nsize, uio);
 478                         if (error)
 479                                 break;
 480                         cpu_mfence();
 481                         rpipe->pipe_buffer.rindex += nsize;
 482                         nread += nsize;
 483
 484                         /*
 485                          * If the FIFO is still over half full just continue
 486                          * and do not try to notify the writer yet.
 487                          */
 488                         if (size - nsize >= (rpipe->pipe_buffer.size >> 1)) {
 489                                 notify_writer = 0;
 490                                 continue;
 491                         }
 492
 493                         /*
 494                          * When the FIFO is less then half full notify any
 495                          * waiting writer.  WANTW can be checked while
 496                          * holding just the rlock.
 497                          */
 498                         notify_writer = 1;
 499                         if ((rpipe->pipe_state & PIPE_WANTW) == 0)
 500                                 continue;
 501                 }
 502
 503                 /*
 504                  * If the "write-side" was blocked we wake it up.  This code
 505                  * is reached either when the buffer is completely emptied
 506                  * or if it becomes more then half-empty.
 507                  *
 508                  * Pipe_state can only be modified if both the rlock and
 509                  * wlock are held.
 510                  */
 511                 if (rpipe->pipe_state & PIPE_WANTW) {
 512                         lwkt_gettoken(&rpipe->pipe_wlock);
 513                         if (rpipe->pipe_state & PIPE_WANTW) {
 514                                 rpipe->pipe_state &= ~PIPE_WANTW;
 515                                 lwkt_reltoken(&rpipe->pipe_wlock);
 516                                 wakeup(rpipe);
 517                         } else {
 518                                 lwkt_reltoken(&rpipe->pipe_wlock);
 519                         }
 520                 }
 521
 522                 /*
 523                  * Pick up our copy loop again if the writer sent data to
 524                  * us while we were messing around.
 525                  *
 526                  * On a SMP box poll up to pipe_delay nanoseconds for new
 527                  * data.  Typically a value of 2000 to 4000 is sufficient
 528                  * to eradicate most IPIs/tsleeps/wakeups when a pipe
 529                  * is used for synchronous communications with small packets,
 530                  * and 8000 or so (8uS) will pipeline large buffer xfers
 531                  * between cpus over a pipe.
 532                  *
 533                  * For synchronous communications a hit means doing a
 534                  * full Awrite-Bread-Bwrite-Aread cycle in less then 2uS,
 535                  * where as miss requiring a tsleep/wakeup sequence
 536                  * will take 7uS or more.
 537                  */
 538                 if (rpipe->pipe_buffer.windex != rpipe->pipe_buffer.rindex)
 539                         continue;
 540
 541 #ifdef _RDTSC_SUPPORTED_
 542                 if (pipe_delay) {
 543                         int64_t tsc_target;
 544                         int good = 0;
 545
 546                         tsc_target = tsc_get_target(pipe_delay);
 547                         while (tsc_test_target(tsc_target) == 0) {
 548                                 if (rpipe->pipe_buffer.windex !=
 549                                     rpipe->pipe_buffer.rindex) {
 550                                         good = 1;
 551                                         break;
 552                                 }
 553                         }
 554                         if (good)
 555                                 continue;
 556                 }
 557 #endif
 558
 559                 /*
 560                  * Detect EOF condition, do not set error.
 561                  */
 562                 if (rpipe->pipe_state & PIPE_REOF)
 563                         break;
 564
 565                 /*
 566                  * Break if some data was read, or if this was a non-blocking
 567                  * read.
 568                  */
 569                 if (nread > 0)
 570                         break;
 571
 572                 if (nbio) {
 573                         error = EAGAIN;
 574                         break;
 575                 }
 576
 577                 /*
 578                  * Last chance, interlock with WANTR.
 579                  */
 580                 lwkt_gettoken(&rpipe->pipe_wlock);
 581                 size = rpipe->pipe_buffer.windex - rpipe->pipe_buffer.rindex;
 582                 if (size) {
 583                         lwkt_reltoken(&rpipe->pipe_wlock);
 584                         continue;
 585                 }
 586
 587                 /*
 588                  * Retest EOF - acquiring a new token can temporarily release
 589                  * tokens already held.
 590                  */
 591                 if (rpipe->pipe_state & PIPE_REOF) {
 592                         lwkt_reltoken(&rpipe->pipe_wlock);
 593                         break;
 594                 }
 595
 596                 /*
 597                  * If there is no more to read in the pipe, reset its
 598                  * pointers to the beginning.  This improves cache hit
 599                  * stats.
 600                  *
 601                  * We need both locks to modify both pointers, and there
 602                  * must also not be a write in progress or the uiomove()
 603                  * in the write might block and temporarily release
 604                  * its wlock, then reacquire and update windex.  We are
 605                  * only serialized against reads, not writes.
 606                  *
 607                  * XXX should we even bother resetting the indices?  It
 608                  *     might actually be more cache efficient not to.
 609                  */
 610                 if (rpipe->pipe_buffer.rindex == rpipe->pipe_buffer.windex &&
 611                     rpipe->pipe_wip == 0) {
 612                         rpipe->pipe_buffer.rindex = 0;
 613                         rpipe->pipe_buffer.windex = 0;
 614                 }
 615
 616                 /*
 617                  * Wait for more data.
 618                  *
 619                  * Pipe_state can only be set if both the rlock and wlock
 620                  * are held.
 621                  */
 622                 rpipe->pipe_state |= PIPE_WANTR;
 623                 tsleep_interlock(rpipe, PCATCH);
 624                 lwkt_reltoken(&rpipe->pipe_wlock);
 625                 error = tsleep(rpipe, PCATCH | PINTERLOCKED, "piperd", 0);
 626                 ++pipe_rblocked_count;
 627                 if (error)
 628                         break;
 629         }
 630         pipe_end_uio(rpipe, &rpipe->pipe_rip);
 631
 632         /*
 633          * Uptime last access time
 634          */
 635         if (error == 0 && nread)
 636                 vfs_timestamp(&rpipe->pipe_atime);
 637
 638         /*
 639          * If we drained the FIFO more then half way then handle
 640          * write blocking hysteresis.
 641          *
 642          * Note that PIPE_WANTW cannot be set by the writer without
 643          * it holding both rlock and wlock, so we can test it
 644          * while holding just rlock.
 645          */
 646         if (notify_writer) {
 647                 /*
 648                  * Synchronous blocking is done on the pipe involved
 649                  */
 650                 if (rpipe->pipe_state & PIPE_WANTW) {
 651                         lwkt_gettoken(&rpipe->pipe_wlock);
 652                         if (rpipe->pipe_state & PIPE_WANTW) {
 653                                 rpipe->pipe_state &= ~PIPE_WANTW;
 654                                 lwkt_reltoken(&rpipe->pipe_wlock);
 655                                 wakeup(rpipe);
 656                         } else {
 657                                 lwkt_reltoken(&rpipe->pipe_wlock);
 658                         }
 659                 }
 660
 661                 /*
 662                  * But we may also have to deal with a kqueue which is
 663                  * stored on the same pipe as its descriptor, so a
 664                  * EVFILT_WRITE event waiting for our side to drain will
 665                  * be on the other side.
 666                  */
 667                 lwkt_gettoken(&wpipe->pipe_wlock);
 668                 pipewakeup(wpipe, 0);
 669                 lwkt_reltoken(&wpipe->pipe_wlock);
 670         }
 671         /*size = rpipe->pipe_buffer.windex - rpipe->pipe_buffer.rindex;*/
 672         lwkt_reltoken(&rpipe->pipe_rlock);
 673
 674         return (error);
 675 }
 676
 677 static int
 678 pipe_write(struct file *fp, struct uio *uio, struct ucred *cred, int fflags)
 679 {
 680         int error;
 681         int orig_resid;
 682         int nbio;
 683         struct pipe *wpipe;
 684         struct pipe *rpipe;
 685         u_int windex;
 686         u_int space;
 687         u_int wcount;
 688         int bigwrite;
 689         int bigcount;
 690
 691         /*
 692          * Writes go to the peer.  The peer will always exist.
 693          */
 694         rpipe = (struct pipe *) fp->f_data;
 695         wpipe = rpipe->pipe_peer;
 696         lwkt_gettoken(&wpipe->pipe_wlock);
 697         if (wpipe->pipe_state & PIPE_WEOF) {
 698                 lwkt_reltoken(&wpipe->pipe_wlock);
 699                 return (EPIPE);
 700         }
 701
 702         /*
 703          * Degenerate case (EPIPE takes prec)
 704          */
 705         if (uio->uio_resid == 0) {
 706                 lwkt_reltoken(&wpipe->pipe_wlock);
 707                 return(0);
 708         }
 709
 710         /*
 711          * Writes are serialized (start_uio must be called with wlock)
 712          */
 713         error = pipe_start_uio(wpipe, &wpipe->pipe_wip);
 714         if (error) {
 715                 lwkt_reltoken(&wpipe->pipe_wlock);
 716                 return (error);
 717         }
 718
 719         if (fflags & O_FBLOCKING)
 720                 nbio = 0;
 721         else if (fflags & O_FNONBLOCKING)
 722                 nbio = 1;
 723         else if (fp->f_flag & O_NONBLOCK)
 724                 nbio = 1;
 725         else
 726                 nbio = 0;
 727
 728         /*
 729          * If it is advantageous to resize the pipe buffer, do
 730          * so.  We are write-serialized so we can block safely.
 731          */
 732         if ((wpipe->pipe_buffer.size <= PIPE_SIZE) &&
 733             (pipe_nbig < pipe_maxbig) &&
 734             wpipe->pipe_wantwcnt > 4 &&
 735             (wpipe->pipe_buffer.rindex == wpipe->pipe_buffer.windex)) {
 736                 /*
 737                  * Recheck after lock.
 738                  */
 739                 lwkt_gettoken(&wpipe->pipe_rlock);
 740                 if ((wpipe->pipe_buffer.size <= PIPE_SIZE) &&
 741                     (pipe_nbig < pipe_maxbig) &&
 742                     (wpipe->pipe_buffer.rindex == wpipe->pipe_buffer.windex)) {
 743                         atomic_add_int(&pipe_nbig, 1);
 744                         if (pipespace(wpipe, BIG_PIPE_SIZE) == 0)
 745                                 ++pipe_bigcount;
 746                         else
 747                                 atomic_subtract_int(&pipe_nbig, 1);
 748                 }
 749                 lwkt_reltoken(&wpipe->pipe_rlock);
 750         }
 751
 752         orig_resid = uio->uio_resid;
 753         wcount = 0;
 754
 755         bigwrite = (uio->uio_resid > 10 * 1024 * 1024);
 756         bigcount = 10;
 757
 758         while (uio->uio_resid) {
 759                 if (wpipe->pipe_state & PIPE_WEOF) {
 760                         error = EPIPE;
 761                         break;
 762                 }
 763
 764                 /*
 765                  * Don't hog the cpu.
 766                  */
 767                 if (bigwrite && --bigcount == 0) {
 768                         lwkt_user_yield();
 769                         bigcount = 10;
 770                         if (CURSIG(curthread->td_lwp)) {
 771                                 error = EINTR;
 772                                 break;
 773                         }
 774                 }
 775
 776                 windex = wpipe->pipe_buffer.windex &
 777                          (wpipe->pipe_buffer.size - 1);
 778                 space = wpipe->pipe_buffer.size -
 779                         (wpipe->pipe_buffer.windex - wpipe->pipe_buffer.rindex);
 780                 cpu_lfence();
 781
 782                 /* Writes of size <= PIPE_BUF must be atomic. */
 783                 if ((space < uio->uio_resid) && (orig_resid <= PIPE_BUF))
 784                         space = 0;
 785
 786                 /*
 787                  * Write to fill, read size handles write hysteresis.  Also
 788                  * additional restrictions can cause select-based non-blocking
 789                  * writes to spin.
 790                  */
 791                 if (space > 0) {
 792                         u_int segsize;
 793
 794                         /*
 795                          * Transfer size is minimum of uio transfer
 796                          * and free space in pipe buffer.
 797                          *
 798                          * Limit each uiocopy to no more then PIPE_SIZE
 799                          * so we can keep the gravy train going on a
 800                          * SMP box.  This doubles the performance for
 801                          * write sizes > 16K.  Otherwise large writes
 802                          * wind up doing an inefficient synchronous
 803                          * ping-pong.
 804                          */
 805                         space = szmin(space, uio->uio_resid);
 806                         if (space > PIPE_SIZE)
 807                                 space = PIPE_SIZE;
 808
 809                         /*
 810                          * First segment to transfer is minimum of
 811                          * transfer size and contiguous space in
 812                          * pipe buffer.  If first segment to transfer
 813                          * is less than the transfer size, we've got
 814                          * a wraparound in the buffer.
 815                          */
 816                         segsize = wpipe->pipe_buffer.size - windex;
 817                         if (segsize > space)
 818                                 segsize = space;
 819
 820                         /*
 821                          * If this is the first loop and the reader is
 822                          * blocked, do a preemptive wakeup of the reader.
 823                          *
 824                          * On SMP the IPI latency plus the wlock interlock
 825                          * on the reader side is the fastest way to get the
 826                          * reader going.  (The scheduler will hard loop on
 827                          * lock tokens).
 828                          *
 829                          * NOTE: We can't clear WANTR here without acquiring
 830                          * the rlock, which we don't want to do here!
 831                          */
 832                         if ((wpipe->pipe_state & PIPE_WANTR))
 833                                 wakeup(wpipe);
 834
 835                         /*
 836                          * Transfer segment, which may include a wrap-around.
 837                          * Update windex to account for both all in one go
 838                          * so the reader can read() the data atomically.
 839                          */
 840                         error = uiomove(&wpipe->pipe_buffer.buffer[windex],
 841                                         segsize, uio);
 842                         if (error == 0 && segsize < space) {
 843                                 segsize = space - segsize;
 844                                 error = uiomove(&wpipe->pipe_buffer.buffer[0],
 845                                                 segsize, uio);
 846                         }
 847                         if (error)
 848                                 break;
 849                         cpu_mfence();
 850                         wpipe->pipe_buffer.windex += space;
 851                         wcount += space;
 852                         continue;
 853                 }
 854
 855                 /*
 856                  * We need both the rlock and the wlock to interlock against
 857                  * the EOF, WANTW, and size checks, and to modify pipe_state.
 858                  *
 859                  * These are token locks so we do not have to worry about
 860                  * deadlocks.
 861                  */
 862                 lwkt_gettoken(&wpipe->pipe_rlock);
 863
 864                 /*
 865                  * If the "read-side" has been blocked, wake it up now
 866                  * and yield to let it drain synchronously rather
 867                  * then block.
 868                  */
 869                 if (wpipe->pipe_state & PIPE_WANTR) {
 870                         wpipe->pipe_state &= ~PIPE_WANTR;
 871                         wakeup(wpipe);
 872                 }
 873
 874                 /*
 875                  * don't block on non-blocking I/O
 876                  */
 877                 if (nbio) {
 878                         lwkt_reltoken(&wpipe->pipe_rlock);
 879                         error = EAGAIN;
 880                         break;
 881                 }
 882
 883                 /*
 884                  * re-test whether we have to block in the writer after
 885                  * acquiring both locks, in case the reader opened up
 886                  * some space.
 887                  */
 888                 space = wpipe->pipe_buffer.size -
 889                         (wpipe->pipe_buffer.windex - wpipe->pipe_buffer.rindex);
 890                 cpu_lfence();
 891                 if ((space < uio->uio_resid) && (orig_resid <= PIPE_BUF))
 892                         space = 0;
 893
 894                 /*
 895                  * Retest EOF - acquiring a new token can temporarily release
 896                  * tokens already held.
 897                  */
 898                 if (wpipe->pipe_state & PIPE_WEOF) {
 899                         lwkt_reltoken(&wpipe->pipe_rlock);
 900                         error = EPIPE;
 901                         break;
 902                 }
 903
 904                 /*
 905                  * We have no more space and have something to offer,
 906                  * wake up select/poll/kq.
 907                  */
 908                 if (space == 0) {
 909                         wpipe->pipe_state |= PIPE_WANTW;
 910                         ++wpipe->pipe_wantwcnt;
 911                         pipewakeup(wpipe, 1);
 912                         if (wpipe->pipe_state & PIPE_WANTW)
 913                                 error = tsleep(wpipe, PCATCH, "pipewr", 0);
 914                         ++pipe_wblocked_count;
 915                 }
 916                 lwkt_reltoken(&wpipe->pipe_rlock);
 917
 918                 /*
 919                  * Break out if we errored or the read side wants us to go
 920                  * away.
 921                  */
 922                 if (error)
 923                         break;
 924                 if (wpipe->pipe_state & PIPE_WEOF) {
 925                         error = EPIPE;
 926                         break;
 927                 }
 928         }
 929         pipe_end_uio(wpipe, &wpipe->pipe_wip);
 930
 931         /*
 932          * If we have put any characters in the buffer, we wake up
 933          * the reader.
 934          *
 935          * Both rlock and wlock are required to be able to modify pipe_state.
 936          */
 937         if (wpipe->pipe_buffer.windex != wpipe->pipe_buffer.rindex) {
 938                 if (wpipe->pipe_state & PIPE_WANTR) {
 939                         lwkt_gettoken(&wpipe->pipe_rlock);
 940                         if (wpipe->pipe_state & PIPE_WANTR) {
 941                                 wpipe->pipe_state &= ~PIPE_WANTR;
 942                                 lwkt_reltoken(&wpipe->pipe_rlock);
 943                                 wakeup(wpipe);
 944                         } else {
 945                                 lwkt_reltoken(&wpipe->pipe_rlock);
 946                         }
 947                 }
 948                 lwkt_gettoken(&wpipe->pipe_rlock);
 949                 pipewakeup(wpipe, 1);
 950                 lwkt_reltoken(&wpipe->pipe_rlock);
 951         }
 952
 953         /*
 954          * Don't return EPIPE if I/O was successful
 955          */
 956         if ((wpipe->pipe_buffer.rindex == wpipe->pipe_buffer.windex) &&
 957             (uio->uio_resid == 0) &&
 958             (error == EPIPE)) {
 959                 error = 0;
 960         }
 961
 962         if (error == 0)
 963                 vfs_timestamp(&wpipe->pipe_mtime);
 964
 965         /*
 966          * We have something to offer,
 967          * wake up select/poll/kq.
 968          */
 969         /*space = wpipe->pipe_buffer.windex - wpipe->pipe_buffer.rindex;*/
 970         lwkt_reltoken(&wpipe->pipe_wlock);
 971         return (error);
 972 }
 973
 974 /*
 975  * we implement a very minimal set of ioctls for compatibility with sockets.
 976  */
 977 static int
 978 pipe_ioctl(struct file *fp, u_long cmd, caddr_t data,
 979            struct ucred *cred, struct sysmsg *msg)
 980 {
 981         struct pipe *mpipe;
 982         int error;
 983
 984         mpipe = (struct pipe *)fp->f_data;
 985
 986         lwkt_gettoken(&mpipe->pipe_rlock);
 987         lwkt_gettoken(&mpipe->pipe_wlock);
 988
 989         switch (cmd) {
 990         case FIOASYNC:
 991                 if (*(int *)data) {
 992                         mpipe->pipe_state |= PIPE_ASYNC;
 993                 } else {
 994                         mpipe->pipe_state &= ~PIPE_ASYNC;
 995                 }
 996                 error = 0;
 997                 break;
 998         case FIONREAD:
 999                 *(int *)data = mpipe->pipe_buffer.windex -
1000                                 mpipe->pipe_buffer.rindex;
1001                 error = 0;
1002                 break;
1003         case FIOSETOWN:
1004                 error = fsetown(*(int *)data, &mpipe->pipe_sigio);
1005                 break;
1006         case FIOGETOWN:
1007                 *(int *)data = fgetown(&mpipe->pipe_sigio);
1008                 error = 0;
1009                 break;
1010         case TIOCSPGRP:
1011                 /* This is deprecated, FIOSETOWN should be used instead. */
1012                 error = fsetown(-(*(int *)data), &mpipe->pipe_sigio);
1013                 break;
1014
1015         case TIOCGPGRP:
1016                 /* This is deprecated, FIOGETOWN should be used instead. */
1017                 *(int *)data = -fgetown(&mpipe->pipe_sigio);
1018                 error = 0;
1019                 break;
1020         default:
1021                 error = ENOTTY;
1022                 break;
1023         }
1024         lwkt_reltoken(&mpipe->pipe_wlock);
1025         lwkt_reltoken(&mpipe->pipe_rlock);
1026
1027         return (error);
1028 }
1029
1030 /*
1031  * MPSAFE
1032  */
1033 static int
1034 pipe_stat(struct file *fp, struct stat *ub, struct ucred *cred)
1035 {
1036         struct pipe *pipe;
1037
1038         pipe = (struct pipe *)fp->f_data;
1039
1040         bzero((caddr_t)ub, sizeof(*ub));
1041         ub->st_mode = S_IFIFO;
1042         ub->st_blksize = pipe->pipe_buffer.size;
1043         ub->st_size = pipe->pipe_buffer.windex - pipe->pipe_buffer.rindex;
1044         ub->st_blocks = (ub->st_size + ub->st_blksize - 1) / ub->st_blksize;
1045         ub->st_atimespec = pipe->pipe_atime;
1046         ub->st_mtimespec = pipe->pipe_mtime;
1047         ub->st_ctimespec = pipe->pipe_ctime;
1048         /*
1049          * Left as 0: st_dev, st_ino, st_nlink, st_uid, st_gid, st_rdev,
1050          * st_flags, st_gen.
1051          * XXX (st_dev, st_ino) should be unique.
1052          */
1053         return (0);
1054 }
1055
1056 static int
1057 pipe_close(struct file *fp)
1058 {
1059         struct pipe *cpipe;
1060
1061         cpipe = (struct pipe *)fp->f_data;
1062         fp->f_ops = &badfileops;
1063         fp->f_data = NULL;
1064         funsetown(&cpipe->pipe_sigio);
1065         pipeclose(cpipe);
1066         return (0);
1067 }
1068
1069 /*
1070  * Shutdown one or both directions of a full-duplex pipe.
1071  */
1072 static int
1073 pipe_shutdown(struct file *fp, int how)
1074 {
1075         struct pipe *rpipe;
1076         struct pipe *wpipe;
1077         int error = EPIPE;
1078
1079         rpipe = (struct pipe *)fp->f_data;
1080         wpipe = rpipe->pipe_peer;
1081
1082         /*
1083          * We modify pipe_state on both pipes, which means we need
1084          * all four tokens!
1085          */
1086         lwkt_gettoken(&rpipe->pipe_rlock);
1087         lwkt_gettoken(&rpipe->pipe_wlock);
1088         lwkt_gettoken(&wpipe->pipe_rlock);
1089         lwkt_gettoken(&wpipe->pipe_wlock);
1090
1091         switch(how) {
1092         case SHUT_RDWR:
1093         case SHUT_RD:
1094                 rpipe->pipe_state |= PIPE_REOF;         /* my reads */
1095                 rpipe->pipe_state |= PIPE_WEOF;         /* peer writes */
1096                 if (rpipe->pipe_state & PIPE_WANTR) {
1097                         rpipe->pipe_state &= ~PIPE_WANTR;
1098                         wakeup(rpipe);
1099                 }
1100                 if (rpipe->pipe_state & PIPE_WANTW) {
1101                         rpipe->pipe_state &= ~PIPE_WANTW;
1102                         wakeup(rpipe);
1103                 }
1104                 error = 0;
1105                 if (how == SHUT_RD)
1106                         break;
1107                 /* fall through */
1108         case SHUT_WR:
1109                 wpipe->pipe_state |= PIPE_REOF;         /* peer reads */
1110                 wpipe->pipe_state |= PIPE_WEOF;         /* my writes */
1111                 if (wpipe->pipe_state & PIPE_WANTR) {
1112                         wpipe->pipe_state &= ~PIPE_WANTR;
1113                         wakeup(wpipe);
1114                 }
1115                 if (wpipe->pipe_state & PIPE_WANTW) {
1116                         wpipe->pipe_state &= ~PIPE_WANTW;
1117                         wakeup(wpipe);
1118                 }
1119                 error = 0;
1120                 break;
1121         }
1122         pipewakeup(rpipe, 1);
1123         pipewakeup(wpipe, 1);
1124
1125         lwkt_reltoken(&wpipe->pipe_wlock);
1126         lwkt_reltoken(&wpipe->pipe_rlock);
1127         lwkt_reltoken(&rpipe->pipe_wlock);
1128         lwkt_reltoken(&rpipe->pipe_rlock);
1129
1130         return (error);
1131 }
1132
1133 static void
1134 pipe_free_kmem(struct pipe *cpipe)
1135 {
1136         if (cpipe->pipe_buffer.buffer != NULL) {
1137                 if (cpipe->pipe_buffer.size > PIPE_SIZE)
1138                         atomic_subtract_int(&pipe_nbig, 1);
1139                 kmem_free(&kernel_map,
1140                         (vm_offset_t)cpipe->pipe_buffer.buffer,
1141                         cpipe->pipe_buffer.size);
1142                 cpipe->pipe_buffer.buffer = NULL;
1143                 cpipe->pipe_buffer.object = NULL;
1144         }
1145 }
1146
1147 /*
1148  * Close the pipe.  The slock must be held to interlock against simultanious
1149  * closes.  The rlock and wlock must be held to adjust the pipe_state.
1150  */
1151 static void
1152 pipeclose(struct pipe *cpipe)
1153 {
1154         globaldata_t gd;
1155         struct pipe *ppipe;
1156
1157         if (cpipe == NULL)
1158                 return;
1159
1160         /*
1161          * The slock may not have been allocated yet (close during
1162          * initialization)
1163          *
1164          * We need both the read and write tokens to modify pipe_state.
1165          */
1166         if (cpipe->pipe_slock)
1167                 lockmgr(cpipe->pipe_slock, LK_EXCLUSIVE);
1168         lwkt_gettoken(&cpipe->pipe_rlock);
1169         lwkt_gettoken(&cpipe->pipe_wlock);
1170
1171         /*
1172          * Set our state, wakeup anyone waiting in select/poll/kq, and
1173          * wakeup anyone blocked on our pipe.
1174          */
1175         cpipe->pipe_state |= PIPE_CLOSED | PIPE_REOF | PIPE_WEOF;
1176         pipewakeup(cpipe, 1);
1177         if (cpipe->pipe_state & (PIPE_WANTR | PIPE_WANTW)) {
1178                 cpipe->pipe_state &= ~(PIPE_WANTR | PIPE_WANTW);
1179                 wakeup(cpipe);
1180         }
1181
1182         /*
1183          * Disconnect from peer.
1184          */
1185         if ((ppipe = cpipe->pipe_peer) != NULL) {
1186                 lwkt_gettoken(&ppipe->pipe_rlock);
1187                 lwkt_gettoken(&ppipe->pipe_wlock);
1188                 ppipe->pipe_state |= PIPE_REOF | PIPE_WEOF;
1189                 pipewakeup(ppipe, 1);
1190                 if (ppipe->pipe_state & (PIPE_WANTR | PIPE_WANTW)) {
1191                         ppipe->pipe_state &= ~(PIPE_WANTR | PIPE_WANTW);
1192                         wakeup(ppipe);
1193                 }
1194                 if (SLIST_FIRST(&ppipe->pipe_kq.ki_note))
1195                         KNOTE(&ppipe->pipe_kq.ki_note, 0);
1196                 lwkt_reltoken(&ppipe->pipe_wlock);
1197                 lwkt_reltoken(&ppipe->pipe_rlock);
1198         }
1199
1200         /*
1201          * If the peer is also closed we can free resources for both
1202          * sides, otherwise we leave our side intact to deal with any
1203          * races (since we only have the slock).
1204          */
1205         if (ppipe && (ppipe->pipe_state & PIPE_CLOSED)) {
1206                 cpipe->pipe_peer = NULL;
1207                 ppipe->pipe_peer = NULL;
1208                 ppipe->pipe_slock = NULL;       /* we will free the slock */
1209                 pipeclose(ppipe);
1210                 ppipe = NULL;
1211         }
1212
1213         lwkt_reltoken(&cpipe->pipe_wlock);
1214         lwkt_reltoken(&cpipe->pipe_rlock);
1215         if (cpipe->pipe_slock)
1216                 lockmgr(cpipe->pipe_slock, LK_RELEASE);
1217
1218         /*
1219          * If we disassociated from our peer we can free resources
1220          */
1221         if (ppipe == NULL) {
1222                 gd = mycpu;
1223                 if (cpipe->pipe_slock) {
1224                         kfree(cpipe->pipe_slock, M_PIPE);
1225                         cpipe->pipe_slock = NULL;
1226                 }
1227                 if (gd->gd_pipeqcount >= pipe_maxcache ||
1228                     cpipe->pipe_buffer.size != PIPE_SIZE
1229                 ) {
1230                         pipe_free_kmem(cpipe);
1231                         kfree(cpipe, M_PIPE);
1232                 } else {
1233                         cpipe->pipe_state = 0;
1234                         cpipe->pipe_peer = gd->gd_pipeq;
1235                         gd->gd_pipeq = cpipe;
1236                         ++gd->gd_pipeqcount;
1237                 }
1238         }
1239 }
1240
1241 static int
1242 pipe_kqfilter(struct file *fp, struct knote *kn)
1243 {
1244         struct pipe *cpipe;
1245
1246         cpipe = (struct pipe *)kn->kn_fp->f_data;
1247
1248         switch (kn->kn_filter) {
1249         case EVFILT_READ:
1250                 kn->kn_fop = &pipe_rfiltops;
1251                 break;
1252         case EVFILT_WRITE:
1253                 kn->kn_fop = &pipe_wfiltops;
1254                 if (cpipe->pipe_peer == NULL) {
1255                         /* other end of pipe has been closed */
1256                         return (EPIPE);
1257                 }
1258                 break;
1259         default:
1260                 return (EOPNOTSUPP);
1261         }
1262         kn->kn_hook = (caddr_t)cpipe;
1263
1264         knote_insert(&cpipe->pipe_kq.ki_note, kn);
1265
1266         return (0);
1267 }
1268
1269 static void
1270 filt_pipedetach(struct knote *kn)
1271 {
1272         struct pipe *cpipe = (struct pipe *)kn->kn_hook;
1273
1274         knote_remove(&cpipe->pipe_kq.ki_note, kn);
1275 }
1276
1277 /*ARGSUSED*/
1278 static int
1279 filt_piperead(struct knote *kn, long hint)
1280 {
1281         struct pipe *rpipe = (struct pipe *)kn->kn_fp->f_data;
1282         int ready = 0;
1283
1284         lwkt_gettoken(&rpipe->pipe_rlock);
1285         lwkt_gettoken(&rpipe->pipe_wlock);
1286
1287         kn->kn_data = rpipe->pipe_buffer.windex - rpipe->pipe_buffer.rindex;
1288
1289         if (rpipe->pipe_state & PIPE_REOF) {
1290                 /*
1291                  * Only set NODATA if all data has been exhausted
1292                  */
1293                 if (kn->kn_data == 0)
1294                         kn->kn_flags |= EV_NODATA;
1295                 kn->kn_flags |= EV_EOF;
1296                 ready = 1;
1297         }
1298
1299         lwkt_reltoken(&rpipe->pipe_wlock);
1300         lwkt_reltoken(&rpipe->pipe_rlock);
1301
1302         if (!ready)
1303                 ready = kn->kn_data > 0;
1304
1305         return (ready);
1306 }
1307
1308 /*ARGSUSED*/
1309 static int
1310 filt_pipewrite(struct knote *kn, long hint)
1311 {
1312         struct pipe *rpipe = (struct pipe *)kn->kn_fp->f_data;
1313         struct pipe *wpipe = rpipe->pipe_peer;
1314         int ready = 0;
1315
1316         kn->kn_data = 0;
1317         if (wpipe == NULL) {
1318                 kn->kn_flags |= (EV_EOF | EV_NODATA);
1319                 return (1);
1320         }
1321
1322         lwkt_gettoken(&wpipe->pipe_rlock);
1323         lwkt_gettoken(&wpipe->pipe_wlock);
1324
1325         if (wpipe->pipe_state & PIPE_WEOF) {
1326                 kn->kn_flags |= (EV_EOF | EV_NODATA);
1327                 ready = 1;
1328         }
1329
1330         if (!ready)
1331                 kn->kn_data = wpipe->pipe_buffer.size -
1332                               (wpipe->pipe_buffer.windex -
1333                                wpipe->pipe_buffer.rindex);
1334
1335         lwkt_reltoken(&wpipe->pipe_wlock);
1336         lwkt_reltoken(&wpipe->pipe_rlock);
1337
1338         if (!ready)
1339                 ready = kn->kn_data >= PIPE_BUF;
1340
1341         return (ready);
1342 }