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 } 1298