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/poll.h>
  40 #include <sys/select.h>
  41 #include <sys/signalvar.h>
  42 #include <sys/sysproto.h>
  43 #include <sys/pipe.h>
  44 #include <sys/vnode.h>
  45 #include <sys/uio.h>
  46 #include <sys/event.h>
  47 #include <sys/globaldata.h>
  48 #include <sys/module.h>
  49 #include <sys/malloc.h>
  50 #include <sys/sysctl.h>
  51 #include <sys/socket.h>
  52
  53 #include <vm/vm.h>
  54 #include <vm/vm_param.h>
  55 #include <sys/lock.h>
  56 #include <vm/vm_object.h>
  57 #include <vm/vm_kern.h>
  58 #include <vm/vm_extern.h>
  59 #include <vm/pmap.h>
  60 #include <vm/vm_map.h>
  61 #include <vm/vm_page.h>
  62 #include <vm/vm_zone.h>
  63
  64 #include <sys/file2.h>
  65
  66 #include <machine/cpufunc.h>
  67
  68 /*
  69  * interfaces to the outside world
  70  */
  71 static int pipe_read (struct file *fp, struct uio *uio,
  72                 struct ucred *cred, int flags);
  73 static int pipe_write (struct file *fp, struct uio *uio,
  74                 struct ucred *cred, int flags);
  75 static int pipe_close (struct file *fp);
  76 static int pipe_shutdown (struct file *fp, int how);
  77 static int pipe_poll (struct file *fp, int events, struct ucred *cred);
  78 static int pipe_kqfilter (struct file *fp, struct knote *kn);
  79 static int pipe_stat (struct file *fp, struct stat *sb, struct ucred *cred);
  80 static int pipe_ioctl (struct file *fp, u_long cmd, caddr_t data, struct ucred *cred);
  81
  82 static struct fileops pipeops = {
  83         .fo_read = pipe_read,
  84         .fo_write = pipe_write,
  85         .fo_ioctl = pipe_ioctl,
  86         .fo_poll = pipe_poll,
  87         .fo_kqfilter = pipe_kqfilter,
  88         .fo_stat = pipe_stat,
  89         .fo_close = pipe_close,
  90         .fo_shutdown = pipe_shutdown
  91 };
  92
  93 static void     filt_pipedetach(struct knote *kn);
  94 static int      filt_piperead(struct knote *kn, long hint);
  95 static int      filt_pipewrite(struct knote *kn, long hint);
  96
  97 static struct filterops pipe_rfiltops =
  98         { 1, NULL, filt_pipedetach, filt_piperead };
  99 static struct filterops pipe_wfiltops =
 100         { 1, NULL, filt_pipedetach, filt_pipewrite };
 101
 102 MALLOC_DEFINE(M_PIPE, "pipe", "pipe structures");
 103
 104 /*
 105  * Default pipe buffer size(s), this can be kind-of large now because pipe
 106  * space is pageable.  The pipe code will try to maintain locality of
 107  * reference for performance reasons, so small amounts of outstanding I/O
 108  * will not wipe the cache.
 109  */
 110 #define MINPIPESIZE (PIPE_SIZE/3)
 111 #define MAXPIPESIZE (2*PIPE_SIZE/3)
 112
 113 /*
 114  * Limit the number of "big" pipes
 115  */
 116 #define LIMITBIGPIPES   64
 117 #define PIPEQ_MAX_CACHE 16      /* per-cpu pipe structure cache */
 118
 119 static int pipe_maxbig = LIMITBIGPIPES;
 120 static int pipe_maxcache = PIPEQ_MAX_CACHE;
 121 static int pipe_bigcount;
 122 static int pipe_nbig;
 123 static int pipe_bcache_alloc;
 124 static int pipe_bkmem_alloc;
 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, maxcache,
 132         CTLFLAG_RW, &pipe_maxcache, 0, "max pipes cached per-cpu");
 133 SYSCTL_INT(_kern_pipe, OID_AUTO, maxbig,
 134         CTLFLAG_RW, &pipe_maxbig, 0, "max number of big pipes");
 135 #if !defined(NO_PIPE_SYSCTL_STATS)
 136 SYSCTL_INT(_kern_pipe, OID_AUTO, bcache_alloc,
 137         CTLFLAG_RW, &pipe_bcache_alloc, 0, "pipe buffer from pcpu cache");
 138 SYSCTL_INT(_kern_pipe, OID_AUTO, bkmem_alloc,
 139         CTLFLAG_RW, &pipe_bkmem_alloc, 0, "pipe buffer from kmem");
 140 #endif
 141
 142 static void pipeclose (struct pipe *cpipe);
 143 static void pipe_free_kmem (struct pipe *cpipe);
 144 static int pipe_create (struct pipe **cpipep);
 145 static __inline int pipelock (struct pipe *cpipe, int catch);
 146 static __inline void pipeunlock (struct pipe *cpipe);
 147 static __inline void pipeselwakeup (struct pipe *cpipe);
 148 static int pipespace (struct pipe *cpipe, int size);
 149
 150 /*
 151  * The pipe system call for the DTYPE_PIPE type of pipes
 152  *
 153  * pipe_ARgs(int dummy)
 154  */
 155
 156 /* ARGSUSED */
 157 int
 158 sys_pipe(struct pipe_args *uap)
 159 {
 160         struct thread *td = curthread;
 161         struct proc *p = td->td_proc;
 162         struct file *rf, *wf;
 163         struct pipe *rpipe, *wpipe;
 164         int fd1, fd2, error;
 165
 166         KKASSERT(p);
 167
 168         rpipe = wpipe = NULL;
 169         if (pipe_create(&rpipe) || pipe_create(&wpipe)) {
 170                 pipeclose(rpipe);
 171                 pipeclose(wpipe);
 172                 return (ENFILE);
 173         }
 174
 175         error = falloc(p, &rf, &fd1);
 176         if (error) {
 177                 pipeclose(rpipe);
 178                 pipeclose(wpipe);
 179                 return (error);
 180         }
 181         uap->sysmsg_fds[0] = fd1;
 182
 183         /*
 184          * Warning: once we've gotten past allocation of the fd for the
 185          * read-side, we can only drop the read side via fdrop() in order
 186          * to avoid races against processes which manage to dup() the read
 187          * side while we are blocked trying to allocate the write side.
 188          */
 189         rf->f_type = DTYPE_PIPE;
 190         rf->f_flag = FREAD | FWRITE;
 191         rf->f_ops = &pipeops;
 192         rf->f_data = rpipe;
 193         error = falloc(p, &wf, &fd2);
 194         if (error) {
 195                 fsetfd(p, NULL, fd1);
 196                 fdrop(rf);
 197                 /* rpipe has been closed by fdrop(). */
 198                 pipeclose(wpipe);
 199                 return (error);
 200         }
 201         wf->f_type = DTYPE_PIPE;
 202         wf->f_flag = FREAD | FWRITE;
 203         wf->f_ops = &pipeops;
 204         wf->f_data = wpipe;
 205         uap->sysmsg_fds[1] = fd2;
 206
 207         rpipe->pipe_peer = wpipe;
 208         wpipe->pipe_peer = rpipe;
 209
 210         fsetfd(p, rf, fd1);
 211         fsetfd(p, wf, fd2);
 212         fdrop(rf);
 213         fdrop(wf);
 214
 215         return (0);
 216 }
 217
 218 /*
 219  * Allocate kva for pipe circular buffer, the space is pageable
 220  * This routine will 'realloc' the size of a pipe safely, if it fails
 221  * it will retain the old buffer.
 222  * If it fails it will return ENOMEM.
 223  */
 224 static int
 225 pipespace(struct pipe *cpipe, int size)
 226 {
 227         struct vm_object *object;
 228         caddr_t buffer;
 229         int npages, error;
 230
 231         npages = round_page(size) / PAGE_SIZE;
 232         object = cpipe->pipe_buffer.object;
 233
 234         /*
 235          * [re]create the object if necessary and reserve space for it
 236          * in the kernel_map.  The object and memory are pageable.  On
 237          * success, free the old resources before assigning the new
 238          * ones.
 239          */
 240         if (object == NULL || object->size != npages) {
 241                 object = vm_object_allocate(OBJT_DEFAULT, npages);
 242                 buffer = (caddr_t)vm_map_min(&kernel_map);
 243
 244                 error = vm_map_find(&kernel_map, object, 0,
 245                                     (vm_offset_t *)&buffer, size,
 246                                     1,
 247                                     VM_MAPTYPE_NORMAL,
 248                                     VM_PROT_ALL, VM_PROT_ALL,
 249                                     0);
 250
 251                 if (error != KERN_SUCCESS) {
 252                         vm_object_deallocate(object);
 253                         return (ENOMEM);
 254                 }
 255                 pipe_free_kmem(cpipe);
 256                 cpipe->pipe_buffer.object = object;
 257                 cpipe->pipe_buffer.buffer = buffer;
 258                 cpipe->pipe_buffer.size = size;
 259                 ++pipe_bkmem_alloc;
 260         } else {
 261                 ++pipe_bcache_alloc;
 262         }
 263         cpipe->pipe_buffer.rindex = 0;
 264         cpipe->pipe_buffer.windex = 0;
 265         return (0);
 266 }
 267
 268 /*
 269  * Initialize and allocate VM and memory for pipe, pulling the pipe from
 270  * our per-cpu cache if possible.  For now make sure it is sized for the
 271  * smaller PIPE_SIZE default.
 272  */
 273 static int
 274 pipe_create(struct pipe **cpipep)
 275 {
 276         globaldata_t gd = mycpu;
 277         struct pipe *cpipe;
 278         int error;
 279
 280         if ((cpipe = gd->gd_pipeq) != NULL) {
 281                 gd->gd_pipeq = cpipe->pipe_peer;
 282                 --gd->gd_pipeqcount;
 283                 cpipe->pipe_peer = NULL;
 284         } else {
 285                 cpipe = kmalloc(sizeof(struct pipe), M_PIPE, M_WAITOK|M_ZERO);
 286         }
 287         *cpipep = cpipe;
 288         if ((error = pipespace(cpipe, PIPE_SIZE)) != 0)
 289                 return (error);
 290         vfs_timestamp(&cpipe->pipe_ctime);
 291         cpipe->pipe_atime = cpipe->pipe_ctime;
 292         cpipe->pipe_mtime = cpipe->pipe_ctime;
 293         return (0);
 294 }
 295
 296
 297 /*
 298  * lock a pipe for I/O, blocking other access
 299  */
 300 static __inline int
 301 pipelock(struct pipe *cpipe, int catch)
 302 {
 303         int error;
 304
 305         while (cpipe->pipe_state & PIPE_LOCK) {
 306                 cpipe->pipe_state |= PIPE_LWANT;
 307                 error = tsleep(cpipe, (catch ? PCATCH : 0), "pipelk", 0);
 308                 if (error != 0)
 309                         return (error);
 310         }
 311         cpipe->pipe_state |= PIPE_LOCK;
 312         return (0);
 313 }
 314
 315 /*
 316  * unlock a pipe I/O lock
 317  */
 318 static __inline void
 319 pipeunlock(struct pipe *cpipe)
 320 {
 321
 322         cpipe->pipe_state &= ~PIPE_LOCK;
 323         if (cpipe->pipe_state & PIPE_LWANT) {
 324                 cpipe->pipe_state &= ~PIPE_LWANT;
 325                 wakeup(cpipe);
 326         }
 327 }
 328
 329 static __inline void
 330 pipeselwakeup(struct pipe *cpipe)
 331 {
 332
 333         if (cpipe->pipe_state & PIPE_SEL) {
 334                 cpipe->pipe_state &= ~PIPE_SEL;
 335                 selwakeup(&cpipe->pipe_sel);
 336         }
 337         if ((cpipe->pipe_state & PIPE_ASYNC) && cpipe->pipe_sigio)
 338                 pgsigio(cpipe->pipe_sigio, SIGIO, 0);
 339         KNOTE(&cpipe->pipe_sel.si_note, 0);
 340 }
 341
 342 /*
 343  * MPALMOSTSAFE (acquires mplock)
 344  */
 345 static int
 346 pipe_read(struct file *fp, struct uio *uio, struct ucred *cred, int fflags)
 347 {
 348         struct pipe *rpipe;
 349         int error;
 350         int nread = 0;
 351         int nbio;
 352         u_int size;     /* total bytes available */
 353         u_int rindex;   /* contiguous bytes available */
 354
 355         get_mplock();
 356         rpipe = (struct pipe *) fp->f_data;
 357         ++rpipe->pipe_busy;
 358         error = pipelock(rpipe, 1);
 359         if (error)
 360                 goto unlocked_error;
 361
 362         if (fflags & O_FBLOCKING)
 363                 nbio = 0;
 364         else if (fflags & O_FNONBLOCKING)
 365                 nbio = 1;
 366         else if (fp->f_flag & O_NONBLOCK)
 367                 nbio = 1;
 368         else
 369                 nbio = 0;
 370
 371         while (uio->uio_resid) {
 372                 size = rpipe->pipe_buffer.windex - rpipe->pipe_buffer.rindex;
 373                 if (size) {
 374                         rindex = rpipe->pipe_buffer.rindex &
 375                                  (rpipe->pipe_buffer.size - 1);
 376                         if (size > rpipe->pipe_buffer.size - rindex)
 377                                 size = rpipe->pipe_buffer.size - rindex;
 378                         if (size > (u_int)uio->uio_resid)
 379                                 size = (u_int)uio->uio_resid;
 380
 381                         error = uiomove(&rpipe->pipe_buffer.buffer[rindex],
 382                                         size, uio);
 383                         if (error)
 384                                 break;
 385                         rpipe->pipe_buffer.rindex += size;
 386
 387                         /*
 388                          * If there is no more to read in the pipe, reset
 389                          * its pointers to the beginning.  This improves
 390                          * cache hit stats.
 391                          */
 392                         if (rpipe->pipe_buffer.rindex ==
 393                             rpipe->pipe_buffer.windex) {
 394                                 rpipe->pipe_buffer.rindex = 0;
 395                                 rpipe->pipe_buffer.windex = 0;
 396                         }
 397                         nread += size;
 398                 } else {
 399                         /*
 400                          * detect EOF condition
 401                          * read returns 0 on EOF, no need to set error
 402                          */
 403                         if (rpipe->pipe_state & PIPE_EOF)
 404                                 break;
 405
 406                         /*
 407                          * If the "write-side" has been blocked, wake it up now.
 408                          */
 409                         if (rpipe->pipe_state & PIPE_WANTW) {
 410                                 rpipe->pipe_state &= ~PIPE_WANTW;
 411                                 wakeup(rpipe);
 412                         }
 413
 414                         /*
 415                          * Break if some data was read.
 416                          */
 417                         if (nread > 0)
 418                                 break;
 419
 420                         /*
 421                          * Unlock the pipe buffer for our remaining
 422                          * processing.  We will either break out with an
 423                          * error or we will sleep and relock to loop.
 424                          */
 425                         pipeunlock(rpipe);
 426
 427                         /*
 428                          * Handle non-blocking mode operation or
 429                          * wait for more data.
 430                          */
 431                         if (nbio) {
 432                                 error = EAGAIN;
 433                         } else {
 434                                 rpipe->pipe_state |= PIPE_WANTR;
 435                                 if ((error = tsleep(rpipe, PCATCH,
 436                                                     "piperd", 0)) == 0) {
 437                                         error = pipelock(rpipe, 1);
 438                                 }
 439                         }
 440                         if (error)
 441                                 goto unlocked_error;
 442                 }
 443         }
 444         pipeunlock(rpipe);
 445
 446         if (error == 0)
 447                 vfs_timestamp(&rpipe->pipe_atime);
 448 unlocked_error:
 449         --rpipe->pipe_busy;
 450
 451         /*
 452          * PIPE_WANT processing only makes sense if pipe_busy is 0.
 453          */
 454         size = rpipe->pipe_buffer.windex - rpipe->pipe_buffer.rindex;
 455
 456         if ((rpipe->pipe_busy == 0) && (rpipe->pipe_state & PIPE_WANT)) {
 457                 rpipe->pipe_state &= ~(PIPE_WANT|PIPE_WANTW);
 458                 wakeup(rpipe);
 459         } else if (size < MINPIPESIZE) {
 460                 /*
 461                  * Handle write blocking hysteresis.
 462                  */
 463                 if (rpipe->pipe_state & PIPE_WANTW) {
 464                         rpipe->pipe_state &= ~PIPE_WANTW;
 465                         wakeup(rpipe);
 466                 }
 467         }
 468
 469         if ((rpipe->pipe_buffer.size - size) >= PIPE_BUF)
 470                 pipeselwakeup(rpipe);
 471         rel_mplock();
 472         return (error);
 473 }
 474
 475 /*
 476  * MPALMOSTSAFE - acquires mplock
 477  */
 478 static int
 479 pipe_write(struct file *fp, struct uio *uio, struct ucred *cred, int fflags)
 480 {
 481         int error = 0;
 482         int orig_resid;
 483         int nbio;
 484         struct pipe *wpipe, *rpipe;
 485         u_int windex;
 486         u_int space;
 487
 488         get_mplock();
 489         rpipe = (struct pipe *) fp->f_data;
 490         wpipe = rpipe->pipe_peer;
 491
 492         /*
 493          * detect loss of pipe read side, issue SIGPIPE if lost.
 494          */
 495         if ((wpipe == NULL) || (wpipe->pipe_state & PIPE_EOF)) {
 496                 rel_mplock();
 497                 return (EPIPE);
 498         }
 499         ++wpipe->pipe_busy;
 500
 501         if (fflags & O_FBLOCKING)
 502                 nbio = 0;
 503         else if (fflags & O_FNONBLOCKING)
 504                 nbio = 1;
 505         else if (fp->f_flag & O_NONBLOCK)
 506                 nbio = 1;
 507         else
 508                 nbio = 0;
 509
 510         /*
 511          * If it is advantageous to resize the pipe buffer, do
 512          * so.
 513          */
 514         if ((uio->uio_resid > PIPE_SIZE) &&
 515             (pipe_nbig < pipe_maxbig) &&
 516             (wpipe->pipe_buffer.size <= PIPE_SIZE) &&
 517             (wpipe->pipe_buffer.rindex == wpipe->pipe_buffer.windex) &&
 518             (error = pipelock(wpipe, 1)) == 0) {
 519                 /*
 520                  * Recheck after lock.
 521                  */
 522                 if ((pipe_nbig < pipe_maxbig) &&
 523                     (wpipe->pipe_buffer.size <= PIPE_SIZE) &&
 524                     (wpipe->pipe_buffer.rindex == wpipe->pipe_buffer.windex)) {
 525                         if (pipespace(wpipe, BIG_PIPE_SIZE) == 0) {
 526                                 ++pipe_bigcount;
 527                                 pipe_nbig++;
 528                         }
 529                 }
 530                 pipeunlock(wpipe);
 531         }
 532
 533         /*
 534          * If an early error occured unbusy and return, waking up any pending
 535          * readers.
 536          */
 537         if (error) {
 538                 --wpipe->pipe_busy;
 539                 if ((wpipe->pipe_busy == 0) &&
 540                     (wpipe->pipe_state & PIPE_WANT)) {
 541                         wpipe->pipe_state &= ~(PIPE_WANT | PIPE_WANTR);
 542                         wakeup(wpipe);
 543                 }
 544                 rel_mplock();
 545                 return(error);
 546         }
 547
 548         KASSERT(wpipe->pipe_buffer.buffer != NULL, ("pipe buffer gone"));
 549
 550         orig_resid = uio->uio_resid;
 551
 552         while (uio->uio_resid) {
 553                 if (wpipe->pipe_state & PIPE_EOF) {
 554                         error = EPIPE;
 555                         break;
 556                 }
 557
 558                 windex = wpipe->pipe_buffer.windex &
 559                          (wpipe->pipe_buffer.size - 1);
 560                 space = wpipe->pipe_buffer.size -
 561                         (wpipe->pipe_buffer.windex - wpipe->pipe_buffer.rindex);
 562
 563                 /* Writes of size <= PIPE_BUF must be atomic. */
 564                 if ((space < uio->uio_resid) && (orig_resid <= PIPE_BUF))
 565                         space = 0;
 566
 567                 /*
 568                  * Write to fill, read size handles write hysteresis.  Also
 569                  * additional restrictions can cause select-based non-blocking
 570                  * writes to spin.
 571                  */
 572                 if (space > 0) {
 573                         if ((error = pipelock(wpipe,1)) == 0) {
 574                                 u_int segsize;
 575
 576                                 /*
 577                                  * If a process blocked in uiomove, our
 578                                  * value for space might be bad.
 579                                  *
 580                                  * XXX will we be ok if the reader has gone
 581                                  * away here?
 582                                  */
 583                                 if (space > (wpipe->pipe_buffer.size -
 584                                              (wpipe->pipe_buffer.windex -
 585                                               wpipe->pipe_buffer.rindex))) {
 586                                         pipeunlock(wpipe);
 587                                         continue;
 588                                 }
 589                                 windex = wpipe->pipe_buffer.windex &
 590                                          (wpipe->pipe_buffer.size - 1);
 591
 592                                 /*
 593                                  * Transfer size is minimum of uio transfer
 594                                  * and free space in pipe buffer.
 595                                  */
 596                                 if (space > (u_int)uio->uio_resid)
 597                                         space = (u_int)uio->uio_resid;
 598
 599                                 /*
 600                                  * First segment to transfer is minimum of
 601                                  * transfer size and contiguous space in
 602                                  * pipe buffer.  If first segment to transfer
 603                                  * is less than the transfer size, we've got
 604                                  * a wraparound in the buffer.
 605                                  */
 606                                 segsize = wpipe->pipe_buffer.size - windex;
 607                                 if (segsize > space)
 608                                         segsize = space;
 609
 610                                 /* Transfer first segment */
 611
 612                                 error = uiomove(
 613                                             &wpipe->pipe_buffer.buffer[windex],
 614                                             segsize, uio);
 615
 616                                 if (error == 0 && segsize < space) {
 617                                         /*
 618                                          * Transfer remaining part now, to
 619                                          * support atomic writes.  Wraparound
 620                                          * happened.
 621                                          */
 622                                         error = uiomove(&wpipe->pipe_buffer.
 623                                                           buffer[0],
 624                                                         space - segsize, uio);
 625                                 }
 626                                 if (error == 0)
 627                                         wpipe->pipe_buffer.windex += space;
 628                                 pipeunlock(wpipe);
 629                         }
 630                         if (error)
 631                                 break;
 632
 633                 } else {
 634                         /*
 635                          * If the "read-side" has been blocked, wake it up now
 636                          * and yield to let it drain synchronously rather
 637                          * then block.
 638                          */
 639                         if (wpipe->pipe_state & PIPE_WANTR) {
 640                                 wpipe->pipe_state &= ~PIPE_WANTR;
 641                                 wakeup(wpipe);
 642                         }
 643
 644                         /*
 645                          * don't block on non-blocking I/O
 646                          */
 647                         if (nbio) {
 648                                 error = EAGAIN;
 649                                 break;
 650                         }
 651
 652                         /*
 653                          * We have no more space and have something to offer,
 654                          * wake up select/poll.
 655                          */
 656                         pipeselwakeup(wpipe);
 657
 658                         wpipe->pipe_state |= PIPE_WANTW;
 659                         error = tsleep(wpipe, PCATCH, "pipewr", 0);
 660                         if (error != 0)
 661                                 break;
 662                         /*
 663                          * If read side wants to go away, we just issue a signal
 664                          * to ourselves.
 665                          */
 666                         if (wpipe->pipe_state & PIPE_EOF) {
 667                                 error = EPIPE;
 668                                 break;
 669                         }
 670                 }
 671         }
 672
 673         --wpipe->pipe_busy;
 674
 675         if ((wpipe->pipe_busy == 0) && (wpipe->pipe_state & PIPE_WANT)) {
 676                 wpipe->pipe_state &= ~(PIPE_WANT | PIPE_WANTR);
 677                 wakeup(wpipe);
 678         } else if (wpipe->pipe_buffer.windex != wpipe->pipe_buffer.rindex) {
 679                 /*
 680                  * If we have put any characters in the buffer, we wake up
 681                  * the reader.
 682                  */
 683                 if (wpipe->pipe_state & PIPE_WANTR) {
 684                         wpipe->pipe_state &= ~PIPE_WANTR;
 685                         wakeup(wpipe);
 686                 }
 687         }
 688
 689         /*
 690          * Don't return EPIPE if I/O was successful
 691          */
 692         if ((wpipe->pipe_buffer.rindex == wpipe->pipe_buffer.windex) &&
 693             (uio->uio_resid == 0) &&
 694             (error == EPIPE)) {
 695                 error = 0;
 696         }
 697
 698         if (error == 0)
 699                 vfs_timestamp(&wpipe->pipe_mtime);
 700
 701         /*
 702          * We have something to offer,
 703          * wake up select/poll.
 704          */
 705         if (wpipe->pipe_buffer.rindex != wpipe->pipe_buffer.windex)
 706                 pipeselwakeup(wpipe);
 707         rel_mplock();
 708         return (error);
 709 }
 710
 711 /*
 712  * MPALMOSTSAFE - acquires mplock
 713  *
 714  * we implement a very minimal set of ioctls for compatibility with sockets.
 715  */
 716 int
 717 pipe_ioctl(struct file *fp, u_long cmd, caddr_t data, struct ucred *cred)
 718 {
 719         struct pipe *mpipe;
 720         int error;
 721
 722         get_mplock();
 723         mpipe = (struct pipe *)fp->f_data;
 724
 725         switch (cmd) {
 726         case FIOASYNC:
 727                 if (*(int *)data) {
 728                         mpipe->pipe_state |= PIPE_ASYNC;
 729                 } else {
 730                         mpipe->pipe_state &= ~PIPE_ASYNC;
 731                 }
 732                 error = 0;
 733                 break;
 734         case FIONREAD:
 735                 *(int *)data = mpipe->pipe_buffer.windex -
 736                                 mpipe->pipe_buffer.rindex;
 737                 error = 0;
 738                 break;
 739         case FIOSETOWN:
 740                 error = fsetown(*(int *)data, &mpipe->pipe_sigio);
 741                 break;
 742         case FIOGETOWN:
 743                 *(int *)data = fgetown(mpipe->pipe_sigio);
 744                 error = 0;
 745                 break;
 746         case TIOCSPGRP:
 747                 /* This is deprecated, FIOSETOWN should be used instead. */
 748                 error = fsetown(-(*(int *)data), &mpipe->pipe_sigio);
 749                 break;
 750
 751         case TIOCGPGRP:
 752                 /* This is deprecated, FIOGETOWN should be used instead. */
 753                 *(int *)data = -fgetown(mpipe->pipe_sigio);
 754                 error = 0;
 755                 break;
 756         default:
 757                 error = ENOTTY;
 758                 break;
 759         }
 760         rel_mplock();
 761         return (error);
 762 }
 763
 764 /*
 765  * MPALMOSTSAFE - acquires mplock
 766  */
 767 int
 768 pipe_poll(struct file *fp, int events, struct ucred *cred)
 769 {
 770         struct pipe *rpipe;
 771         struct pipe *wpipe;
 772         int revents = 0;
 773         u_int space;
 774
 775         get_mplock();
 776         rpipe = (struct pipe *)fp->f_data;
 777         wpipe = rpipe->pipe_peer;
 778         if (events & (POLLIN | POLLRDNORM)) {
 779                 if ((rpipe->pipe_buffer.windex != rpipe->pipe_buffer.rindex) ||
 780                     (rpipe->pipe_state & PIPE_EOF)) {
 781                         revents |= events & (POLLIN | POLLRDNORM);
 782                 }
 783         }
 784
 785         if (events & (POLLOUT | POLLWRNORM)) {
 786                 if (wpipe == NULL || (wpipe->pipe_state & PIPE_EOF)) {
 787                         revents |= events & (POLLOUT | POLLWRNORM);
 788                 } else {
 789                         space = wpipe->pipe_buffer.windex -
 790                                 wpipe->pipe_buffer.rindex;
 791                         space = wpipe->pipe_buffer.size - space;
 792                         if (space >= PIPE_BUF)
 793                                 revents |= events & (POLLOUT | POLLWRNORM);
 794                 }
 795         }
 796
 797         if ((rpipe->pipe_state & PIPE_EOF) ||
 798             (wpipe == NULL) ||
 799             (wpipe->pipe_state & PIPE_EOF))
 800                 revents |= POLLHUP;
 801
 802         if (revents == 0) {
 803                 if (events & (POLLIN | POLLRDNORM)) {
 804                         selrecord(curthread, &rpipe->pipe_sel);
 805                         rpipe->pipe_state |= PIPE_SEL;
 806                 }
 807
 808                 if (events & (POLLOUT | POLLWRNORM)) {
 809                         selrecord(curthread, &wpipe->pipe_sel);
 810                         wpipe->pipe_state |= PIPE_SEL;
 811                 }
 812         }
 813         rel_mplock();
 814         return (revents);
 815 }
 816
 817 /*
 818  * MPALMOSTSAFE - acquires mplock
 819  */
 820 static int
 821 pipe_stat(struct file *fp, struct stat *ub, struct ucred *cred)
 822 {
 823         struct pipe *pipe;
 824
 825         get_mplock();
 826         pipe = (struct pipe *)fp->f_data;
 827
 828         bzero((caddr_t)ub, sizeof(*ub));
 829         ub->st_mode = S_IFIFO;
 830         ub->st_blksize = pipe->pipe_buffer.size;
 831         ub->st_size = pipe->pipe_buffer.windex - pipe->pipe_buffer.rindex;
 832         ub->st_blocks = (ub->st_size + ub->st_blksize - 1) / ub->st_blksize;
 833         ub->st_atimespec = pipe->pipe_atime;
 834         ub->st_mtimespec = pipe->pipe_mtime;
 835         ub->st_ctimespec = pipe->pipe_ctime;
 836         /*
 837          * Left as 0: st_dev, st_ino, st_nlink, st_uid, st_gid, st_rdev,
 838          * st_flags, st_gen.
 839          * XXX (st_dev, st_ino) should be unique.
 840          */
 841         rel_mplock();
 842         return (0);
 843 }
 844
 845 /*
 846  * MPALMOSTSAFE - acquires mplock
 847  */
 848 static int
 849 pipe_close(struct file *fp)
 850 {
 851         struct pipe *cpipe;
 852
 853         get_mplock();
 854         cpipe = (struct pipe *)fp->f_data;
 855         fp->f_ops = &badfileops;
 856         fp->f_data = NULL;
 857         funsetown(cpipe->pipe_sigio);
 858         pipeclose(cpipe);
 859         rel_mplock();
 860         return (0);
 861 }
 862
 863 /*
 864  * Shutdown one or both directions of a full-duplex pipe.
 865  *
 866  * MPALMOSTSAFE - acquires mplock
 867  */
 868 static int
 869 pipe_shutdown(struct file *fp, int how)
 870 {
 871         struct pipe *rpipe;
 872         struct pipe *wpipe;
 873         int error = EPIPE;
 874
 875         get_mplock();
 876         rpipe = (struct pipe *)fp->f_data;
 877
 878         switch(how) {
 879         case SHUT_RDWR:
 880         case SHUT_RD:
 881                 if (rpipe) {
 882                         rpipe->pipe_state |= PIPE_EOF;
 883                         pipeselwakeup(rpipe);
 884                         if (rpipe->pipe_busy)
 885                                 wakeup(rpipe);
 886                         error = 0;
 887                 }
 888                 if (how == SHUT_RD)
 889                         break;
 890                 /* fall through */
 891         case SHUT_WR:
 892                 if (rpipe && (wpipe = rpipe->pipe_peer) != NULL) {
 893                         wpipe->pipe_state |= PIPE_EOF;
 894                         pipeselwakeup(wpipe);
 895                         if (wpipe->pipe_busy)
 896                                 wakeup(wpipe);
 897                         error = 0;
 898                 }
 899         }
 900         rel_mplock();
 901         return (error);
 902 }
 903
 904 static void
 905 pipe_free_kmem(struct pipe *cpipe)
 906 {
 907         if (cpipe->pipe_buffer.buffer != NULL) {
 908                 if (cpipe->pipe_buffer.size > PIPE_SIZE)
 909                         --pipe_nbig;
 910                 kmem_free(&kernel_map,
 911                         (vm_offset_t)cpipe->pipe_buffer.buffer,
 912                         cpipe->pipe_buffer.size);
 913                 cpipe->pipe_buffer.buffer = NULL;
 914                 cpipe->pipe_buffer.object = NULL;
 915         }
 916 }
 917
 918 /*
 919  * shutdown the pipe
 920  */
 921 static void
 922 pipeclose(struct pipe *cpipe)
 923 {
 924         globaldata_t gd;
 925         struct pipe *ppipe;
 926
 927         if (cpipe == NULL)
 928                 return;
 929
 930         pipeselwakeup(cpipe);
 931
 932         /*
 933          * If the other side is blocked, wake it up saying that
 934          * we want to close it down.
 935          */
 936         while (cpipe->pipe_busy) {
 937                 wakeup(cpipe);
 938                 cpipe->pipe_state |= PIPE_WANT | PIPE_EOF;
 939                 tsleep(cpipe, 0, "pipecl", 0);
 940         }
 941
 942         /*
 943          * Disconnect from peer
 944          */
 945         if ((ppipe = cpipe->pipe_peer) != NULL) {
 946                 pipeselwakeup(ppipe);
 947
 948                 ppipe->pipe_state |= PIPE_EOF;
 949                 wakeup(ppipe);
 950                 KNOTE(&ppipe->pipe_sel.si_note, 0);
 951                 ppipe->pipe_peer = NULL;
 952         }
 953
 954         if (cpipe->pipe_kva) {
 955                 pmap_qremove(cpipe->pipe_kva, XIO_INTERNAL_PAGES);
 956                 kmem_free(&kernel_map, cpipe->pipe_kva, XIO_INTERNAL_SIZE);
 957                 cpipe->pipe_kva = 0;
 958         }
 959
 960         /*
 961          * free or cache resources
 962          */
 963         gd = mycpu;
 964         if (gd->gd_pipeqcount >= pipe_maxcache ||
 965             cpipe->pipe_buffer.size != PIPE_SIZE
 966         ) {
 967                 pipe_free_kmem(cpipe);
 968                 kfree(cpipe, M_PIPE);
 969         } else {
 970                 cpipe->pipe_state = 0;
 971                 cpipe->pipe_busy = 0;
 972                 cpipe->pipe_peer = gd->gd_pipeq;
 973                 gd->gd_pipeq = cpipe;
 974                 ++gd->gd_pipeqcount;
 975         }
 976 }
 977
 978 /*
 979  * MPALMOSTSAFE - acquires mplock
 980  */
 981 static int
 982 pipe_kqfilter(struct file *fp, struct knote *kn)
 983 {
 984         struct pipe *cpipe;
 985
 986         get_mplock();
 987         cpipe = (struct pipe *)kn->kn_fp->f_data;
 988
 989         switch (kn->kn_filter) {
 990         case EVFILT_READ:
 991                 kn->kn_fop = &pipe_rfiltops;
 992                 break;
 993         case EVFILT_WRITE:
 994                 kn->kn_fop = &pipe_wfiltops;
 995                 cpipe = cpipe->pipe_peer;
 996                 if (cpipe == NULL) {
 997                         /* other end of pipe has been closed */
 998                         rel_mplock();
 999                         return (EPIPE);
1000                 }
1001                 break;
1002         default:
1003                 return (1);
1004         }
1005         kn->kn_hook = (caddr_t)cpipe;
1006
1007         SLIST_INSERT_HEAD(&cpipe->pipe_sel.si_note, kn, kn_selnext);
1008         rel_mplock();
1009         return (0);
1010 }
1011
1012 static void
1013 filt_pipedetach(struct knote *kn)
1014 {
1015         struct pipe *cpipe = (struct pipe *)kn->kn_hook;
1016
1017         SLIST_REMOVE(&cpipe->pipe_sel.si_note, kn, knote, kn_selnext);
1018 }
1019
1020 /*ARGSUSED*/
1021 static int
1022 filt_piperead(struct knote *kn, long hint)
1023 {
1024         struct pipe *rpipe = (struct pipe *)kn->kn_fp->f_data;
1025         struct pipe *wpipe = rpipe->pipe_peer;
1026
1027         kn->kn_data = rpipe->pipe_buffer.windex - rpipe->pipe_buffer.rindex;
1028
1029         if ((rpipe->pipe_state & PIPE_EOF) ||
1030             (wpipe == NULL) || (wpipe->pipe_state & PIPE_EOF)) {
1031                 kn->kn_flags |= EV_EOF;
1032                 return (1);
1033         }
1034         return (kn->kn_data > 0);
1035 }
1036
1037 /*ARGSUSED*/
1038 static int
1039 filt_pipewrite(struct knote *kn, long hint)
1040 {
1041         struct pipe *rpipe = (struct pipe *)kn->kn_fp->f_data;
1042         struct pipe *wpipe = rpipe->pipe_peer;
1043         u_int32_t space;
1044
1045         if ((wpipe == NULL) || (wpipe->pipe_state & PIPE_EOF)) {
1046                 kn->kn_data = 0;
1047                 kn->kn_flags |= EV_EOF;
1048                 return (1);
1049         }
1050         space = wpipe->pipe_buffer.windex -
1051                 wpipe->pipe_buffer.rindex;
1052         space = wpipe->pipe_buffer.size - space;
1053         kn->kn_data = space;
1054         return (kn->kn_data >= PIPE_BUF);
1055 }