sys_pipe.c revision 91362
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: head/sys/kern/sys_pipe.c 91362 2002-02-27 07:35:59Z alfred $ 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 29/* 30 * This code has two modes of operation, a small write mode and a large 31 * write mode. The small write mode acts like conventional pipes with 32 * a kernel buffer. If the buffer is less than PIPE_MINDIRECT, then the 33 * "normal" pipe buffering is done. If the buffer is between PIPE_MINDIRECT 34 * and PIPE_SIZE in size, it is fully mapped and wired into the kernel, and 35 * the receiving process can copy it directly from the pages in the sending 36 * process. 37 * 38 * If the sending process receives a signal, it is possible that it will 39 * go away, and certainly its address space can change, because control 40 * is returned back to the user-mode side. In that case, the pipe code 41 * arranges to copy the buffer supplied by the user process, to a pageable 42 * kernel buffer, and the receiving process will grab the data from the 43 * pageable kernel buffer. Since signals don't happen all that often, 44 * the copy operation is normally eliminated. 45 * 46 * The constant PIPE_MINDIRECT is chosen to make sure that buffering will 47 * happen for small transfers so that the system will not spend all of 48 * its time context switching. PIPE_SIZE is constrained by the 49 * amount of kernel virtual memory. 50 */ 51 52#include <sys/param.h> 53#include <sys/systm.h> 54#include <sys/fcntl.h> 55#include <sys/file.h> 56#include <sys/filedesc.h> 57#include <sys/filio.h> 58#include <sys/lock.h> 59#include <sys/mutex.h> 60#include <sys/ttycom.h> 61#include <sys/stat.h> 62#include <sys/poll.h> 63#include <sys/selinfo.h> 64#include <sys/signalvar.h> 65#include <sys/sysproto.h> 66#include <sys/pipe.h> 67#include <sys/proc.h> 68#include <sys/vnode.h> 69#include <sys/uio.h> 70#include <sys/event.h> 71 72#include <vm/vm.h> 73#include <vm/vm_param.h> 74#include <vm/vm_object.h> 75#include <vm/vm_kern.h> 76#include <vm/vm_extern.h> 77#include <vm/pmap.h> 78#include <vm/vm_map.h> 79#include <vm/vm_page.h> 80#include <vm/vm_zone.h> 81 82/* 83 * Use this define if you want to disable *fancy* VM things. Expect an 84 * approx 30% decrease in transfer rate. This could be useful for 85 * NetBSD or OpenBSD. 86 */ 87/* #define PIPE_NODIRECT */ 88 89/* 90 * interfaces to the outside world 91 */ 92static int pipe_read __P((struct file *fp, struct uio *uio, 93 struct ucred *cred, int flags, struct thread *td)); 94static int pipe_write __P((struct file *fp, struct uio *uio, 95 struct ucred *cred, int flags, struct thread *td)); 96static int pipe_close __P((struct file *fp, struct thread *td)); 97static int pipe_poll __P((struct file *fp, int events, struct ucred *cred, 98 struct thread *td)); 99static int pipe_kqfilter __P((struct file *fp, struct knote *kn)); 100static int pipe_stat __P((struct file *fp, struct stat *sb, struct thread *td)); 101static int pipe_ioctl __P((struct file *fp, u_long cmd, caddr_t data, struct thread *td)); 102 103static struct fileops pipeops = { 104 pipe_read, pipe_write, pipe_ioctl, pipe_poll, pipe_kqfilter, 105 pipe_stat, pipe_close 106}; 107 108static void filt_pipedetach(struct knote *kn); 109static int filt_piperead(struct knote *kn, long hint); 110static int filt_pipewrite(struct knote *kn, long hint); 111 112static struct filterops pipe_rfiltops = 113 { 1, NULL, filt_pipedetach, filt_piperead }; 114static struct filterops pipe_wfiltops = 115 { 1, NULL, filt_pipedetach, filt_pipewrite }; 116 117#define PIPE_GET_GIANT(pipe) \ 118 do { \ 119 PIPE_UNLOCK(wpipe); \ 120 mtx_lock(&Giant); \ 121 } while (0) 122 123#define PIPE_DROP_GIANT(pipe) \ 124 do { \ 125 mtx_unlock(&Giant); \ 126 PIPE_LOCK(wpipe); \ 127 } while (0) 128 129/* 130 * Default pipe buffer size(s), this can be kind-of large now because pipe 131 * space is pageable. The pipe code will try to maintain locality of 132 * reference for performance reasons, so small amounts of outstanding I/O 133 * will not wipe the cache. 134 */ 135#define MINPIPESIZE (PIPE_SIZE/3) 136#define MAXPIPESIZE (2*PIPE_SIZE/3) 137 138/* 139 * Maximum amount of kva for pipes -- this is kind-of a soft limit, but 140 * is there so that on large systems, we don't exhaust it. 141 */ 142#define MAXPIPEKVA (8*1024*1024) 143 144/* 145 * Limit for direct transfers, we cannot, of course limit 146 * the amount of kva for pipes in general though. 147 */ 148#define LIMITPIPEKVA (16*1024*1024) 149 150/* 151 * Limit the number of "big" pipes 152 */ 153#define LIMITBIGPIPES 32 154static int nbigpipe; 155 156static int amountpipekva; 157 158static void pipeclose __P((struct pipe *cpipe)); 159static void pipe_free_kmem __P((struct pipe *cpipe)); 160static int pipe_create __P((struct pipe **cpipep)); 161static __inline int pipelock __P((struct pipe *cpipe, int catch)); 162static __inline void pipeunlock __P((struct pipe *cpipe)); 163static __inline void pipeselwakeup __P((struct pipe *cpipe)); 164#ifndef PIPE_NODIRECT 165static int pipe_build_write_buffer __P((struct pipe *wpipe, struct uio *uio)); 166static void pipe_destroy_write_buffer __P((struct pipe *wpipe)); 167static int pipe_direct_write __P((struct pipe *wpipe, struct uio *uio)); 168static void pipe_clone_write_buffer __P((struct pipe *wpipe)); 169#endif 170static int pipespace __P((struct pipe *cpipe, int size)); 171 172static vm_zone_t pipe_zone; 173 174/* 175 * The pipe system call for the DTYPE_PIPE type of pipes 176 */ 177 178/* ARGSUSED */ 179int 180pipe(td, uap) 181 struct thread *td; 182 struct pipe_args /* { 183 int dummy; 184 } */ *uap; 185{ 186 struct filedesc *fdp = td->td_proc->p_fd; 187 struct file *rf, *wf; 188 struct pipe *rpipe, *wpipe; 189 int fd, error; 190 191 /* XXX: SYSINIT this! */ 192 if (pipe_zone == NULL) 193 pipe_zone = zinit("PIPE", sizeof(struct pipe), 0, 0, 4); 194 195 rpipe = wpipe = NULL; 196 if (pipe_create(&rpipe) || pipe_create(&wpipe)) { 197 pipeclose(rpipe); 198 pipeclose(wpipe); 199 return (ENFILE); 200 } 201 202 rpipe->pipe_state |= PIPE_DIRECTOK; 203 wpipe->pipe_state |= PIPE_DIRECTOK; 204 205 error = falloc(td, &rf, &fd); 206 if (error) { 207 pipeclose(rpipe); 208 pipeclose(wpipe); 209 return (error); 210 } 211 fhold(rf); 212 td->td_retval[0] = fd; 213 214 /* 215 * Warning: once we've gotten past allocation of the fd for the 216 * read-side, we can only drop the read side via fdrop() in order 217 * to avoid races against processes which manage to dup() the read 218 * side while we are blocked trying to allocate the write side. 219 */ 220 FILE_LOCK(rf); 221 rf->f_flag = FREAD | FWRITE; 222 rf->f_type = DTYPE_PIPE; 223 rf->f_data = (caddr_t)rpipe; 224 rf->f_ops = &pipeops; 225 FILE_UNLOCK(rf); 226 error = falloc(td, &wf, &fd); 227 if (error) { 228 FILEDESC_LOCK(fdp); 229 if (fdp->fd_ofiles[td->td_retval[0]] == rf) { 230 fdp->fd_ofiles[td->td_retval[0]] = NULL; 231 FILEDESC_UNLOCK(fdp); 232 fdrop(rf, td); 233 } else 234 FILEDESC_UNLOCK(fdp); 235 fdrop(rf, td); 236 /* rpipe has been closed by fdrop(). */ 237 pipeclose(wpipe); 238 return (error); 239 } 240 FILE_LOCK(wf); 241 wf->f_flag = FREAD | FWRITE; 242 wf->f_type = DTYPE_PIPE; 243 wf->f_data = (caddr_t)wpipe; 244 wf->f_ops = &pipeops; 245 FILE_UNLOCK(wf); 246 td->td_retval[1] = fd; 247 rpipe->pipe_peer = wpipe; 248 wpipe->pipe_peer = rpipe; 249 rpipe->pipe_mtxp = wpipe->pipe_mtxp = mtx_pool_alloc(); 250 fdrop(rf, td); 251 252 return (0); 253} 254 255/* 256 * Allocate kva for pipe circular buffer, the space is pageable 257 * This routine will 'realloc' the size of a pipe safely, if it fails 258 * it will retain the old buffer. 259 * If it fails it will return ENOMEM. 260 */ 261static int 262pipespace(cpipe, size) 263 struct pipe *cpipe; 264 int size; 265{ 266 struct vm_object *object; 267 caddr_t buffer; 268 int npages, error; 269 270 GIANT_REQUIRED; 271 272 npages = round_page(size)/PAGE_SIZE; 273 /* 274 * Create an object, I don't like the idea of paging to/from 275 * kernel_object. 276 * XXX -- minor change needed here for NetBSD/OpenBSD VM systems. 277 */ 278 object = vm_object_allocate(OBJT_DEFAULT, npages); 279 buffer = (caddr_t) vm_map_min(kernel_map); 280 281 /* 282 * Insert the object into the kernel map, and allocate kva for it. 283 * The map entry is, by default, pageable. 284 * XXX -- minor change needed here for NetBSD/OpenBSD VM systems. 285 */ 286 error = vm_map_find(kernel_map, object, 0, 287 (vm_offset_t *) &buffer, size, 1, 288 VM_PROT_ALL, VM_PROT_ALL, 0); 289 290 if (error != KERN_SUCCESS) { 291 vm_object_deallocate(object); 292 return (ENOMEM); 293 } 294 295 /* free old resources if we're resizing */ 296 pipe_free_kmem(cpipe); 297 cpipe->pipe_buffer.object = object; 298 cpipe->pipe_buffer.buffer = buffer; 299 cpipe->pipe_buffer.size = size; 300 cpipe->pipe_buffer.in = 0; 301 cpipe->pipe_buffer.out = 0; 302 cpipe->pipe_buffer.cnt = 0; 303 amountpipekva += cpipe->pipe_buffer.size; 304 return (0); 305} 306 307/* 308 * initialize and allocate VM and memory for pipe 309 */ 310static int 311pipe_create(cpipep) 312 struct pipe **cpipep; 313{ 314 struct pipe *cpipe; 315 int error; 316 317 *cpipep = zalloc(pipe_zone); 318 if (*cpipep == NULL) 319 return (ENOMEM); 320 321 cpipe = *cpipep; 322 323 /* so pipespace()->pipe_free_kmem() doesn't follow junk pointer */ 324 cpipe->pipe_buffer.object = NULL; 325#ifndef PIPE_NODIRECT 326 cpipe->pipe_map.kva = NULL; 327#endif 328 /* 329 * protect so pipeclose() doesn't follow a junk pointer 330 * if pipespace() fails. 331 */ 332 bzero(&cpipe->pipe_sel, sizeof(cpipe->pipe_sel)); 333 cpipe->pipe_state = 0; 334 cpipe->pipe_peer = NULL; 335 cpipe->pipe_busy = 0; 336 337#ifndef PIPE_NODIRECT 338 /* 339 * pipe data structure initializations to support direct pipe I/O 340 */ 341 cpipe->pipe_map.cnt = 0; 342 cpipe->pipe_map.kva = 0; 343 cpipe->pipe_map.pos = 0; 344 cpipe->pipe_map.npages = 0; 345 /* cpipe->pipe_map.ms[] = invalid */ 346#endif 347 348 error = pipespace(cpipe, PIPE_SIZE); 349 if (error) 350 return (error); 351 352 vfs_timestamp(&cpipe->pipe_ctime); 353 cpipe->pipe_atime = cpipe->pipe_ctime; 354 cpipe->pipe_mtime = cpipe->pipe_ctime; 355 356 return (0); 357} 358 359 360/* 361 * lock a pipe for I/O, blocking other access 362 */ 363static __inline int 364pipelock(cpipe, catch) 365 struct pipe *cpipe; 366 int catch; 367{ 368 int error; 369 370 PIPE_LOCK_ASSERT(cpipe, MA_OWNED); 371 while (cpipe->pipe_state & PIPE_LOCKFL) { 372 cpipe->pipe_state |= PIPE_LWANT; 373 error = msleep(cpipe, PIPE_MTX(cpipe), 374 catch ? (PRIBIO | PCATCH) : PRIBIO, 375 "pipelk", 0); 376 if (error != 0) 377 return (error); 378 } 379 cpipe->pipe_state |= PIPE_LOCKFL; 380 return (0); 381} 382 383/* 384 * unlock a pipe I/O lock 385 */ 386static __inline void 387pipeunlock(cpipe) 388 struct pipe *cpipe; 389{ 390 391 PIPE_LOCK_ASSERT(cpipe, MA_OWNED); 392 cpipe->pipe_state &= ~PIPE_LOCKFL; 393 if (cpipe->pipe_state & PIPE_LWANT) { 394 cpipe->pipe_state &= ~PIPE_LWANT; 395 wakeup(cpipe); 396 } 397} 398 399static __inline void 400pipeselwakeup(cpipe) 401 struct pipe *cpipe; 402{ 403 404 if (cpipe->pipe_state & PIPE_SEL) { 405 cpipe->pipe_state &= ~PIPE_SEL; 406 selwakeup(&cpipe->pipe_sel); 407 } 408 if ((cpipe->pipe_state & PIPE_ASYNC) && cpipe->pipe_sigio) 409 pgsigio(cpipe->pipe_sigio, SIGIO, 0); 410 KNOTE(&cpipe->pipe_sel.si_note, 0); 411} 412 413/* ARGSUSED */ 414static int 415pipe_read(fp, uio, cred, flags, td) 416 struct file *fp; 417 struct uio *uio; 418 struct ucred *cred; 419 struct thread *td; 420 int flags; 421{ 422 struct pipe *rpipe = (struct pipe *) fp->f_data; 423 int error; 424 int nread = 0; 425 u_int size; 426 427 PIPE_LOCK(rpipe); 428 ++rpipe->pipe_busy; 429 error = pipelock(rpipe, 1); 430 if (error) 431 goto unlocked_error; 432 433 while (uio->uio_resid) { 434 /* 435 * normal pipe buffer receive 436 */ 437 if (rpipe->pipe_buffer.cnt > 0) { 438 size = rpipe->pipe_buffer.size - rpipe->pipe_buffer.out; 439 if (size > rpipe->pipe_buffer.cnt) 440 size = rpipe->pipe_buffer.cnt; 441 if (size > (u_int) uio->uio_resid) 442 size = (u_int) uio->uio_resid; 443 444 PIPE_UNLOCK(rpipe); 445 error = uiomove(&rpipe->pipe_buffer.buffer[rpipe->pipe_buffer.out], 446 size, uio); 447 PIPE_LOCK(rpipe); 448 if (error) 449 break; 450 451 rpipe->pipe_buffer.out += size; 452 if (rpipe->pipe_buffer.out >= rpipe->pipe_buffer.size) 453 rpipe->pipe_buffer.out = 0; 454 455 rpipe->pipe_buffer.cnt -= size; 456 457 /* 458 * If there is no more to read in the pipe, reset 459 * its pointers to the beginning. This improves 460 * cache hit stats. 461 */ 462 if (rpipe->pipe_buffer.cnt == 0) { 463 rpipe->pipe_buffer.in = 0; 464 rpipe->pipe_buffer.out = 0; 465 } 466 nread += size; 467#ifndef PIPE_NODIRECT 468 /* 469 * Direct copy, bypassing a kernel buffer. 470 */ 471 } else if ((size = rpipe->pipe_map.cnt) && 472 (rpipe->pipe_state & PIPE_DIRECTW)) { 473 caddr_t va; 474 if (size > (u_int) uio->uio_resid) 475 size = (u_int) uio->uio_resid; 476 477 va = (caddr_t) rpipe->pipe_map.kva + 478 rpipe->pipe_map.pos; 479 PIPE_UNLOCK(rpipe); 480 error = uiomove(va, size, uio); 481 PIPE_LOCK(rpipe); 482 if (error) 483 break; 484 nread += size; 485 rpipe->pipe_map.pos += size; 486 rpipe->pipe_map.cnt -= size; 487 if (rpipe->pipe_map.cnt == 0) { 488 rpipe->pipe_state &= ~PIPE_DIRECTW; 489 wakeup(rpipe); 490 } 491#endif 492 } else { 493 /* 494 * detect EOF condition 495 * read returns 0 on EOF, no need to set error 496 */ 497 if (rpipe->pipe_state & PIPE_EOF) 498 break; 499 500 /* 501 * If the "write-side" has been blocked, wake it up now. 502 */ 503 if (rpipe->pipe_state & PIPE_WANTW) { 504 rpipe->pipe_state &= ~PIPE_WANTW; 505 wakeup(rpipe); 506 } 507 508 /* 509 * Break if some data was read. 510 */ 511 if (nread > 0) 512 break; 513 514 /* 515 * Unlock the pipe buffer for our remaining processing. We 516 * will either break out with an error or we will sleep and 517 * relock to loop. 518 */ 519 pipeunlock(rpipe); 520 521 /* 522 * Handle non-blocking mode operation or 523 * wait for more data. 524 */ 525 if (fp->f_flag & FNONBLOCK) { 526 error = EAGAIN; 527 } else { 528 rpipe->pipe_state |= PIPE_WANTR; 529 if ((error = msleep(rpipe, PIPE_MTX(rpipe), 530 PRIBIO | PCATCH, 531 "piperd", 0)) == 0) 532 error = pipelock(rpipe, 1); 533 } 534 if (error) 535 goto unlocked_error; 536 } 537 } 538 pipeunlock(rpipe); 539 540 /* XXX: should probably do this before getting any locks. */ 541 if (error == 0) 542 vfs_timestamp(&rpipe->pipe_atime); 543unlocked_error: 544 --rpipe->pipe_busy; 545 546 /* 547 * PIPE_WANT processing only makes sense if pipe_busy is 0. 548 */ 549 if ((rpipe->pipe_busy == 0) && (rpipe->pipe_state & PIPE_WANT)) { 550 rpipe->pipe_state &= ~(PIPE_WANT|PIPE_WANTW); 551 wakeup(rpipe); 552 } else if (rpipe->pipe_buffer.cnt < MINPIPESIZE) { 553 /* 554 * Handle write blocking hysteresis. 555 */ 556 if (rpipe->pipe_state & PIPE_WANTW) { 557 rpipe->pipe_state &= ~PIPE_WANTW; 558 wakeup(rpipe); 559 } 560 } 561 562 if ((rpipe->pipe_buffer.size - rpipe->pipe_buffer.cnt) >= PIPE_BUF) 563 pipeselwakeup(rpipe); 564 565 PIPE_UNLOCK(rpipe); 566 return (error); 567} 568 569#ifndef PIPE_NODIRECT 570/* 571 * Map the sending processes' buffer into kernel space and wire it. 572 * This is similar to a physical write operation. 573 */ 574static int 575pipe_build_write_buffer(wpipe, uio) 576 struct pipe *wpipe; 577 struct uio *uio; 578{ 579 u_int size; 580 int i; 581 vm_offset_t addr, endaddr, paddr; 582 583 GIANT_REQUIRED; 584 585 size = (u_int) uio->uio_iov->iov_len; 586 if (size > wpipe->pipe_buffer.size) 587 size = wpipe->pipe_buffer.size; 588 589 endaddr = round_page((vm_offset_t)uio->uio_iov->iov_base + size); 590 addr = trunc_page((vm_offset_t)uio->uio_iov->iov_base); 591 for (i = 0; addr < endaddr; addr += PAGE_SIZE, i++) { 592 vm_page_t m; 593 594 if (vm_fault_quick((caddr_t)addr, VM_PROT_READ) < 0 || 595 (paddr = pmap_kextract(addr)) == 0) { 596 int j; 597 598 for (j = 0; j < i; j++) 599 vm_page_unwire(wpipe->pipe_map.ms[j], 1); 600 return (EFAULT); 601 } 602 603 m = PHYS_TO_VM_PAGE(paddr); 604 vm_page_wire(m); 605 wpipe->pipe_map.ms[i] = m; 606 } 607 608/* 609 * set up the control block 610 */ 611 wpipe->pipe_map.npages = i; 612 wpipe->pipe_map.pos = 613 ((vm_offset_t) uio->uio_iov->iov_base) & PAGE_MASK; 614 wpipe->pipe_map.cnt = size; 615 616/* 617 * and map the buffer 618 */ 619 if (wpipe->pipe_map.kva == 0) { 620 /* 621 * We need to allocate space for an extra page because the 622 * address range might (will) span pages at times. 623 */ 624 wpipe->pipe_map.kva = kmem_alloc_pageable(kernel_map, 625 wpipe->pipe_buffer.size + PAGE_SIZE); 626 amountpipekva += wpipe->pipe_buffer.size + PAGE_SIZE; 627 } 628 pmap_qenter(wpipe->pipe_map.kva, wpipe->pipe_map.ms, 629 wpipe->pipe_map.npages); 630 631/* 632 * and update the uio data 633 */ 634 635 uio->uio_iov->iov_len -= size; 636 uio->uio_iov->iov_base += size; 637 if (uio->uio_iov->iov_len == 0) 638 uio->uio_iov++; 639 uio->uio_resid -= size; 640 uio->uio_offset += size; 641 return (0); 642} 643 644/* 645 * unmap and unwire the process buffer 646 */ 647static void 648pipe_destroy_write_buffer(wpipe) 649 struct pipe *wpipe; 650{ 651 int i; 652 653 GIANT_REQUIRED; 654 655 if (wpipe->pipe_map.kva) { 656 pmap_qremove(wpipe->pipe_map.kva, wpipe->pipe_map.npages); 657 658 if (amountpipekva > MAXPIPEKVA) { 659 vm_offset_t kva = wpipe->pipe_map.kva; 660 wpipe->pipe_map.kva = 0; 661 kmem_free(kernel_map, kva, 662 wpipe->pipe_buffer.size + PAGE_SIZE); 663 amountpipekva -= wpipe->pipe_buffer.size + PAGE_SIZE; 664 } 665 } 666 for (i = 0; i < wpipe->pipe_map.npages; i++) 667 vm_page_unwire(wpipe->pipe_map.ms[i], 1); 668} 669 670/* 671 * In the case of a signal, the writing process might go away. This 672 * code copies the data into the circular buffer so that the source 673 * pages can be freed without loss of data. 674 */ 675static void 676pipe_clone_write_buffer(wpipe) 677 struct pipe *wpipe; 678{ 679 int size; 680 int pos; 681 682 PIPE_LOCK_ASSERT(wpipe, MA_OWNED); 683 size = wpipe->pipe_map.cnt; 684 pos = wpipe->pipe_map.pos; 685 bcopy((caddr_t) wpipe->pipe_map.kva + pos, 686 (caddr_t) wpipe->pipe_buffer.buffer, size); 687 688 wpipe->pipe_buffer.in = size; 689 wpipe->pipe_buffer.out = 0; 690 wpipe->pipe_buffer.cnt = size; 691 wpipe->pipe_state &= ~PIPE_DIRECTW; 692 693 pipe_destroy_write_buffer(wpipe); 694} 695 696/* 697 * This implements the pipe buffer write mechanism. Note that only 698 * a direct write OR a normal pipe write can be pending at any given time. 699 * If there are any characters in the pipe buffer, the direct write will 700 * be deferred until the receiving process grabs all of the bytes from 701 * the pipe buffer. Then the direct mapping write is set-up. 702 */ 703static int 704pipe_direct_write(wpipe, uio) 705 struct pipe *wpipe; 706 struct uio *uio; 707{ 708 int error; 709 710retry: 711 PIPE_LOCK_ASSERT(wpipe, MA_OWNED); 712 while (wpipe->pipe_state & PIPE_DIRECTW) { 713 if (wpipe->pipe_state & PIPE_WANTR) { 714 wpipe->pipe_state &= ~PIPE_WANTR; 715 wakeup(wpipe); 716 } 717 wpipe->pipe_state |= PIPE_WANTW; 718 error = msleep(wpipe, PIPE_MTX(wpipe), 719 PRIBIO | PCATCH, "pipdww", 0); 720 if (error) 721 goto error1; 722 if (wpipe->pipe_state & PIPE_EOF) { 723 error = EPIPE; 724 goto error1; 725 } 726 } 727 wpipe->pipe_map.cnt = 0; /* transfer not ready yet */ 728 if (wpipe->pipe_buffer.cnt > 0) { 729 if (wpipe->pipe_state & PIPE_WANTR) { 730 wpipe->pipe_state &= ~PIPE_WANTR; 731 wakeup(wpipe); 732 } 733 734 wpipe->pipe_state |= PIPE_WANTW; 735 error = msleep(wpipe, PIPE_MTX(wpipe), 736 PRIBIO | PCATCH, "pipdwc", 0); 737 if (error) 738 goto error1; 739 if (wpipe->pipe_state & PIPE_EOF) { 740 error = EPIPE; 741 goto error1; 742 } 743 goto retry; 744 } 745 746 wpipe->pipe_state |= PIPE_DIRECTW; 747 748 PIPE_GET_GIANT(wpipe); 749 error = pipe_build_write_buffer(wpipe, uio); 750 PIPE_DROP_GIANT(wpipe); 751 if (error) { 752 wpipe->pipe_state &= ~PIPE_DIRECTW; 753 goto error1; 754 } 755 756 error = 0; 757 while (!error && (wpipe->pipe_state & PIPE_DIRECTW)) { 758 if (wpipe->pipe_state & PIPE_EOF) { 759 pipelock(wpipe, 0); 760 PIPE_GET_GIANT(wpipe); 761 pipe_destroy_write_buffer(wpipe); 762 PIPE_DROP_GIANT(wpipe); 763 pipeunlock(wpipe); 764 pipeselwakeup(wpipe); 765 error = EPIPE; 766 goto error1; 767 } 768 if (wpipe->pipe_state & PIPE_WANTR) { 769 wpipe->pipe_state &= ~PIPE_WANTR; 770 wakeup(wpipe); 771 } 772 pipeselwakeup(wpipe); 773 error = msleep(wpipe, PIPE_MTX(wpipe), PRIBIO | PCATCH, 774 "pipdwt", 0); 775 } 776 777 pipelock(wpipe,0); 778 if (wpipe->pipe_state & PIPE_DIRECTW) { 779 /* 780 * this bit of trickery substitutes a kernel buffer for 781 * the process that might be going away. 782 */ 783 pipe_clone_write_buffer(wpipe); 784 } else { 785 pipe_destroy_write_buffer(wpipe); 786 } 787 pipeunlock(wpipe); 788 return (error); 789 790error1: 791 wakeup(wpipe); 792 return (error); 793} 794#endif 795 796static int 797pipe_write(fp, uio, cred, flags, td) 798 struct file *fp; 799 struct uio *uio; 800 struct ucred *cred; 801 struct thread *td; 802 int flags; 803{ 804 int error = 0; 805 int orig_resid; 806 struct pipe *wpipe, *rpipe; 807 808 rpipe = (struct pipe *) fp->f_data; 809 wpipe = rpipe->pipe_peer; 810 811 PIPE_LOCK(wpipe); 812 /* 813 * detect loss of pipe read side, issue SIGPIPE if lost. 814 */ 815 if ((wpipe == NULL) || (wpipe->pipe_state & PIPE_EOF)) { 816 PIPE_UNLOCK(wpipe); 817 return (EPIPE); 818 } 819 ++wpipe->pipe_busy; 820 821 /* 822 * If it is advantageous to resize the pipe buffer, do 823 * so. 824 */ 825 if ((uio->uio_resid > PIPE_SIZE) && 826 (nbigpipe < LIMITBIGPIPES) && 827 (wpipe->pipe_state & PIPE_DIRECTW) == 0 && 828 (wpipe->pipe_buffer.size <= PIPE_SIZE) && 829 (wpipe->pipe_buffer.cnt == 0)) { 830 831 if ((error = pipelock(wpipe,1)) == 0) { 832 PIPE_GET_GIANT(wpipe); 833 if (pipespace(wpipe, BIG_PIPE_SIZE) == 0) 834 nbigpipe++; 835 PIPE_DROP_GIANT(wpipe); 836 pipeunlock(wpipe); 837 } 838 } 839 840 /* 841 * If an early error occured unbusy and return, waking up any pending 842 * readers. 843 */ 844 if (error) { 845 --wpipe->pipe_busy; 846 if ((wpipe->pipe_busy == 0) && 847 (wpipe->pipe_state & PIPE_WANT)) { 848 wpipe->pipe_state &= ~(PIPE_WANT | PIPE_WANTR); 849 wakeup(wpipe); 850 } 851 PIPE_UNLOCK(wpipe); 852 return(error); 853 } 854 855 KASSERT(wpipe->pipe_buffer.buffer != NULL, ("pipe buffer gone")); 856 857 orig_resid = uio->uio_resid; 858 859 while (uio->uio_resid) { 860 int space; 861 862#ifndef PIPE_NODIRECT 863 /* 864 * If the transfer is large, we can gain performance if 865 * we do process-to-process copies directly. 866 * If the write is non-blocking, we don't use the 867 * direct write mechanism. 868 * 869 * The direct write mechanism will detect the reader going 870 * away on us. 871 */ 872 if ((uio->uio_iov->iov_len >= PIPE_MINDIRECT) && 873 (fp->f_flag & FNONBLOCK) == 0 && 874 (wpipe->pipe_map.kva || (amountpipekva < LIMITPIPEKVA)) && 875 (uio->uio_iov->iov_len >= PIPE_MINDIRECT)) { 876 error = pipe_direct_write( wpipe, uio); 877 if (error) 878 break; 879 continue; 880 } 881#endif 882 883 /* 884 * Pipe buffered writes cannot be coincidental with 885 * direct writes. We wait until the currently executing 886 * direct write is completed before we start filling the 887 * pipe buffer. We break out if a signal occurs or the 888 * reader goes away. 889 */ 890 retrywrite: 891 while (wpipe->pipe_state & PIPE_DIRECTW) { 892 if (wpipe->pipe_state & PIPE_WANTR) { 893 wpipe->pipe_state &= ~PIPE_WANTR; 894 wakeup(wpipe); 895 } 896 error = msleep(wpipe, PIPE_MTX(wpipe), PRIBIO | PCATCH, 897 "pipbww", 0); 898 if (wpipe->pipe_state & PIPE_EOF) 899 break; 900 if (error) 901 break; 902 } 903 if (wpipe->pipe_state & PIPE_EOF) { 904 error = EPIPE; 905 break; 906 } 907 908 space = wpipe->pipe_buffer.size - wpipe->pipe_buffer.cnt; 909 910 /* Writes of size <= PIPE_BUF must be atomic. */ 911 if ((space < uio->uio_resid) && (orig_resid <= PIPE_BUF)) 912 space = 0; 913 914 if (space > 0 && (wpipe->pipe_buffer.cnt < PIPE_SIZE)) { 915 if ((error = pipelock(wpipe,1)) == 0) { 916 int size; /* Transfer size */ 917 int segsize; /* first segment to transfer */ 918 919 /* 920 * It is possible for a direct write to 921 * slip in on us... handle it here... 922 */ 923 if (wpipe->pipe_state & PIPE_DIRECTW) { 924 pipeunlock(wpipe); 925 goto retrywrite; 926 } 927 /* 928 * If a process blocked in uiomove, our 929 * value for space might be bad. 930 * 931 * XXX will we be ok if the reader has gone 932 * away here? 933 */ 934 if (space > wpipe->pipe_buffer.size - 935 wpipe->pipe_buffer.cnt) { 936 pipeunlock(wpipe); 937 goto retrywrite; 938 } 939 940 /* 941 * Transfer size is minimum of uio transfer 942 * and free space in pipe buffer. 943 */ 944 if (space > uio->uio_resid) 945 size = uio->uio_resid; 946 else 947 size = space; 948 /* 949 * First segment to transfer is minimum of 950 * transfer size and contiguous space in 951 * pipe buffer. If first segment to transfer 952 * is less than the transfer size, we've got 953 * a wraparound in the buffer. 954 */ 955 segsize = wpipe->pipe_buffer.size - 956 wpipe->pipe_buffer.in; 957 if (segsize > size) 958 segsize = size; 959 960 /* Transfer first segment */ 961 962 PIPE_UNLOCK(wpipe); 963 error = uiomove(&wpipe->pipe_buffer.buffer[wpipe->pipe_buffer.in], 964 segsize, uio); 965 PIPE_LOCK(wpipe); 966 967 if (error == 0 && segsize < size) { 968 /* 969 * Transfer remaining part now, to 970 * support atomic writes. Wraparound 971 * happened. 972 */ 973 if (wpipe->pipe_buffer.in + segsize != 974 wpipe->pipe_buffer.size) 975 panic("Expected pipe buffer wraparound disappeared"); 976 977 PIPE_UNLOCK(wpipe); 978 error = uiomove(&wpipe->pipe_buffer.buffer[0], 979 size - segsize, uio); 980 PIPE_LOCK(wpipe); 981 } 982 if (error == 0) { 983 wpipe->pipe_buffer.in += size; 984 if (wpipe->pipe_buffer.in >= 985 wpipe->pipe_buffer.size) { 986 if (wpipe->pipe_buffer.in != size - segsize + wpipe->pipe_buffer.size) 987 panic("Expected wraparound bad"); 988 wpipe->pipe_buffer.in = size - segsize; 989 } 990 991 wpipe->pipe_buffer.cnt += size; 992 if (wpipe->pipe_buffer.cnt > wpipe->pipe_buffer.size) 993 panic("Pipe buffer overflow"); 994 995 } 996 pipeunlock(wpipe); 997 } 998 if (error) 999 break; 1000 1001 } else { 1002 /* 1003 * If the "read-side" has been blocked, wake it up now. 1004 */ 1005 if (wpipe->pipe_state & PIPE_WANTR) { 1006 wpipe->pipe_state &= ~PIPE_WANTR; 1007 wakeup(wpipe); 1008 } 1009 1010 /* 1011 * don't block on non-blocking I/O 1012 */ 1013 if (fp->f_flag & FNONBLOCK) { 1014 error = EAGAIN; 1015 break; 1016 } 1017 1018 /* 1019 * We have no more space and have something to offer, 1020 * wake up select/poll. 1021 */ 1022 pipeselwakeup(wpipe); 1023 1024 wpipe->pipe_state |= PIPE_WANTW; 1025 error = msleep(wpipe, PIPE_MTX(wpipe), 1026 PRIBIO | PCATCH, "pipewr", 0); 1027 if (error != 0) 1028 break; 1029 /* 1030 * If read side wants to go away, we just issue a signal 1031 * to ourselves. 1032 */ 1033 if (wpipe->pipe_state & PIPE_EOF) { 1034 error = EPIPE; 1035 break; 1036 } 1037 } 1038 } 1039 1040 --wpipe->pipe_busy; 1041 1042 if ((wpipe->pipe_busy == 0) && (wpipe->pipe_state & PIPE_WANT)) { 1043 wpipe->pipe_state &= ~(PIPE_WANT | PIPE_WANTR); 1044 wakeup(wpipe); 1045 } else if (wpipe->pipe_buffer.cnt > 0) { 1046 /* 1047 * If we have put any characters in the buffer, we wake up 1048 * the reader. 1049 */ 1050 if (wpipe->pipe_state & PIPE_WANTR) { 1051 wpipe->pipe_state &= ~PIPE_WANTR; 1052 wakeup(wpipe); 1053 } 1054 } 1055 1056 /* 1057 * Don't return EPIPE if I/O was successful 1058 */ 1059 if ((wpipe->pipe_buffer.cnt == 0) && 1060 (uio->uio_resid == 0) && 1061 (error == EPIPE)) { 1062 error = 0; 1063 } 1064 1065 if (error == 0) 1066 vfs_timestamp(&wpipe->pipe_mtime); 1067 1068 /* 1069 * We have something to offer, 1070 * wake up select/poll. 1071 */ 1072 if (wpipe->pipe_buffer.cnt) 1073 pipeselwakeup(wpipe); 1074 1075 PIPE_UNLOCK(wpipe); 1076 return (error); 1077} 1078 1079/* 1080 * we implement a very minimal set of ioctls for compatibility with sockets. 1081 */ 1082int 1083pipe_ioctl(fp, cmd, data, td) 1084 struct file *fp; 1085 u_long cmd; 1086 caddr_t data; 1087 struct thread *td; 1088{ 1089 struct pipe *mpipe = (struct pipe *)fp->f_data; 1090 1091 switch (cmd) { 1092 1093 case FIONBIO: 1094 return (0); 1095 1096 case FIOASYNC: 1097 PIPE_LOCK(mpipe); 1098 if (*(int *)data) { 1099 mpipe->pipe_state |= PIPE_ASYNC; 1100 } else { 1101 mpipe->pipe_state &= ~PIPE_ASYNC; 1102 } 1103 PIPE_UNLOCK(mpipe); 1104 return (0); 1105 1106 case FIONREAD: 1107 PIPE_LOCK(mpipe); 1108 if (mpipe->pipe_state & PIPE_DIRECTW) 1109 *(int *)data = mpipe->pipe_map.cnt; 1110 else 1111 *(int *)data = mpipe->pipe_buffer.cnt; 1112 PIPE_UNLOCK(mpipe); 1113 return (0); 1114 1115 case FIOSETOWN: 1116 return (fsetown(*(int *)data, &mpipe->pipe_sigio)); 1117 1118 case FIOGETOWN: 1119 *(int *)data = fgetown(mpipe->pipe_sigio); 1120 return (0); 1121 1122 /* This is deprecated, FIOSETOWN should be used instead. */ 1123 case TIOCSPGRP: 1124 return (fsetown(-(*(int *)data), &mpipe->pipe_sigio)); 1125 1126 /* This is deprecated, FIOGETOWN should be used instead. */ 1127 case TIOCGPGRP: 1128 *(int *)data = -fgetown(mpipe->pipe_sigio); 1129 return (0); 1130 1131 } 1132 return (ENOTTY); 1133} 1134 1135int 1136pipe_poll(fp, events, cred, td) 1137 struct file *fp; 1138 int events; 1139 struct ucred *cred; 1140 struct thread *td; 1141{ 1142 struct pipe *rpipe = (struct pipe *)fp->f_data; 1143 struct pipe *wpipe; 1144 int revents = 0; 1145 1146 wpipe = rpipe->pipe_peer; 1147 PIPE_LOCK(rpipe); 1148 if (events & (POLLIN | POLLRDNORM)) 1149 if ((rpipe->pipe_state & PIPE_DIRECTW) || 1150 (rpipe->pipe_buffer.cnt > 0) || 1151 (rpipe->pipe_state & PIPE_EOF)) 1152 revents |= events & (POLLIN | POLLRDNORM); 1153 1154 if (events & (POLLOUT | POLLWRNORM)) 1155 if (wpipe == NULL || (wpipe->pipe_state & PIPE_EOF) || 1156 (((wpipe->pipe_state & PIPE_DIRECTW) == 0) && 1157 (wpipe->pipe_buffer.size - wpipe->pipe_buffer.cnt) >= PIPE_BUF)) 1158 revents |= events & (POLLOUT | POLLWRNORM); 1159 1160 if ((rpipe->pipe_state & PIPE_EOF) || 1161 (wpipe == NULL) || 1162 (wpipe->pipe_state & PIPE_EOF)) 1163 revents |= POLLHUP; 1164 1165 if (revents == 0) { 1166 if (events & (POLLIN | POLLRDNORM)) { 1167 selrecord(td, &rpipe->pipe_sel); 1168 rpipe->pipe_state |= PIPE_SEL; 1169 } 1170 1171 if (events & (POLLOUT | POLLWRNORM)) { 1172 selrecord(td, &wpipe->pipe_sel); 1173 wpipe->pipe_state |= PIPE_SEL; 1174 } 1175 } 1176 PIPE_UNLOCK(rpipe); 1177 1178 return (revents); 1179} 1180 1181static int 1182pipe_stat(fp, ub, td) 1183 struct file *fp; 1184 struct stat *ub; 1185 struct thread *td; 1186{ 1187 struct pipe *pipe = (struct pipe *)fp->f_data; 1188 1189 bzero((caddr_t)ub, sizeof(*ub)); 1190 ub->st_mode = S_IFIFO; 1191 ub->st_blksize = pipe->pipe_buffer.size; 1192 ub->st_size = pipe->pipe_buffer.cnt; 1193 ub->st_blocks = (ub->st_size + ub->st_blksize - 1) / ub->st_blksize; 1194 ub->st_atimespec = pipe->pipe_atime; 1195 ub->st_mtimespec = pipe->pipe_mtime; 1196 ub->st_ctimespec = pipe->pipe_ctime; 1197 ub->st_uid = fp->f_cred->cr_uid; 1198 ub->st_gid = fp->f_cred->cr_gid; 1199 /* 1200 * Left as 0: st_dev, st_ino, st_nlink, st_rdev, st_flags, st_gen. 1201 * XXX (st_dev, st_ino) should be unique. 1202 */ 1203 return (0); 1204} 1205 1206/* ARGSUSED */ 1207static int 1208pipe_close(fp, td) 1209 struct file *fp; 1210 struct thread *td; 1211{ 1212 struct pipe *cpipe = (struct pipe *)fp->f_data; 1213 1214 fp->f_ops = &badfileops; 1215 fp->f_data = NULL; 1216 funsetown(cpipe->pipe_sigio); 1217 pipeclose(cpipe); 1218 return (0); 1219} 1220 1221static void 1222pipe_free_kmem(cpipe) 1223 struct pipe *cpipe; 1224{ 1225 GIANT_REQUIRED; 1226 1227 if (cpipe->pipe_buffer.buffer != NULL) { 1228 if (cpipe->pipe_buffer.size > PIPE_SIZE) 1229 --nbigpipe; 1230 amountpipekva -= cpipe->pipe_buffer.size; 1231 kmem_free(kernel_map, 1232 (vm_offset_t)cpipe->pipe_buffer.buffer, 1233 cpipe->pipe_buffer.size); 1234 cpipe->pipe_buffer.buffer = NULL; 1235 } 1236#ifndef PIPE_NODIRECT 1237 if (cpipe->pipe_map.kva != NULL) { 1238 amountpipekva -= cpipe->pipe_buffer.size + PAGE_SIZE; 1239 kmem_free(kernel_map, 1240 cpipe->pipe_map.kva, 1241 cpipe->pipe_buffer.size + PAGE_SIZE); 1242 cpipe->pipe_map.cnt = 0; 1243 cpipe->pipe_map.kva = 0; 1244 cpipe->pipe_map.pos = 0; 1245 cpipe->pipe_map.npages = 0; 1246 } 1247#endif 1248} 1249 1250/* 1251 * shutdown the pipe 1252 */ 1253static void 1254pipeclose(cpipe) 1255 struct pipe *cpipe; 1256{ 1257 struct pipe *ppipe; 1258 1259 if (cpipe) { 1260 PIPE_LOCK(cpipe); 1261 1262 pipeselwakeup(cpipe); 1263 1264 /* 1265 * If the other side is blocked, wake it up saying that 1266 * we want to close it down. 1267 */ 1268 while (cpipe->pipe_busy) { 1269 wakeup(cpipe); 1270 cpipe->pipe_state |= PIPE_WANT | PIPE_EOF; 1271 msleep(cpipe, PIPE_MTX(cpipe), PRIBIO, "pipecl", 0); 1272 } 1273 1274 /* 1275 * Disconnect from peer 1276 */ 1277 if ((ppipe = cpipe->pipe_peer) != NULL) { 1278 pipeselwakeup(ppipe); 1279 1280 ppipe->pipe_state |= PIPE_EOF; 1281 wakeup(ppipe); 1282 KNOTE(&ppipe->pipe_sel.si_note, 0); 1283 ppipe->pipe_peer = NULL; 1284 } 1285 /* 1286 * free resources 1287 */ 1288 PIPE_UNLOCK(cpipe); 1289 mtx_lock(&Giant); 1290 pipe_free_kmem(cpipe); 1291 zfree(pipe_zone, cpipe); 1292 mtx_unlock(&Giant); 1293 } 1294} 1295 1296/*ARGSUSED*/ 1297static int 1298pipe_kqfilter(struct file *fp, struct knote *kn) 1299{ 1300 struct pipe *cpipe; 1301 1302 cpipe = (struct pipe *)kn->kn_fp->f_data; 1303 switch (kn->kn_filter) { 1304 case EVFILT_READ: 1305 kn->kn_fop = &pipe_rfiltops; 1306 break; 1307 case EVFILT_WRITE: 1308 kn->kn_fop = &pipe_wfiltops; 1309 cpipe = cpipe->pipe_peer; 1310 break; 1311 default: 1312 return (1); 1313 } 1314 kn->kn_hook = (caddr_t)cpipe; 1315 1316 SLIST_INSERT_HEAD(&cpipe->pipe_sel.si_note, kn, kn_selnext); 1317 return (0); 1318} 1319 1320static void 1321filt_pipedetach(struct knote *kn) 1322{ 1323 struct pipe *cpipe = (struct pipe *)kn->kn_hook; 1324 1325 SLIST_REMOVE(&cpipe->pipe_sel.si_note, kn, knote, kn_selnext); 1326} 1327 1328/*ARGSUSED*/ 1329static int 1330filt_piperead(struct knote *kn, long hint) 1331{ 1332 struct pipe *rpipe = (struct pipe *)kn->kn_fp->f_data; 1333 struct pipe *wpipe = rpipe->pipe_peer; 1334 1335 kn->kn_data = rpipe->pipe_buffer.cnt; 1336 if ((kn->kn_data == 0) && (rpipe->pipe_state & PIPE_DIRECTW)) 1337 kn->kn_data = rpipe->pipe_map.cnt; 1338 1339 if ((rpipe->pipe_state & PIPE_EOF) || 1340 (wpipe == NULL) || (wpipe->pipe_state & PIPE_EOF)) { 1341 kn->kn_flags |= EV_EOF; 1342 return (1); 1343 } 1344 return (kn->kn_data > 0); 1345} 1346 1347/*ARGSUSED*/ 1348static int 1349filt_pipewrite(struct knote *kn, long hint) 1350{ 1351 struct pipe *rpipe = (struct pipe *)kn->kn_fp->f_data; 1352 struct pipe *wpipe = rpipe->pipe_peer; 1353 1354 if ((wpipe == NULL) || (wpipe->pipe_state & PIPE_EOF)) { 1355 kn->kn_data = 0; 1356 kn->kn_flags |= EV_EOF; 1357 return (1); 1358 } 1359 kn->kn_data = wpipe->pipe_buffer.size - wpipe->pipe_buffer.cnt; 1360 if (wpipe->pipe_state & PIPE_DIRECTW) 1361 kn->kn_data = 0; 1362 1363 return (kn->kn_data >= PIPE_BUF); 1364} 1365