1 /* 2 * Copyright (c) 1996 John S. Dyson 3 * All rights reserved. 4 * 5 * Redistribution and use in source and binary forms, with or without 6 * modification, are permitted provided that the following conditions 7 * are met: 8 * 1. Redistributions of source code must retain the above copyright 9 * notice immediately at the beginning of the file, without modification, 10 * this list of conditions, and the following disclaimer. 11 * 2. Redistributions in binary form must reproduce the above copyright 12 * notice, this list of conditions and the following disclaimer in the 13 * documentation and/or other materials provided with the distribution. 14 * 3. Absolutely no warranty of function or purpose is made by the author 15 * John S. Dyson. 16 * 4. Modifications may be freely made to this file if the above conditions 17 * are met. 18 * 19 * $FreeBSD: src/sys/kern/sys_pipe.c,v 1.60.2.13 2002/08/05 15:05:15 des Exp $ 20 */ 21 22 /* 23 * This file contains a high-performance replacement for the socket-based 24 * pipes scheme originally used in FreeBSD/4.4Lite. It does not support 25 * all features of sockets, but does do everything that pipes normally 26 * do. 27 */ 28 #include <sys/param.h> 29 #include <sys/systm.h> 30 #include <sys/kernel.h> 31 #include <sys/proc.h> 32 #include <sys/fcntl.h> 33 #include <sys/file.h> 34 #include <sys/filedesc.h> 35 #include <sys/filio.h> 36 #include <sys/ttycom.h> 37 #include <sys/stat.h> 38 #include <sys/signalvar.h> 39 #include <sys/sysproto.h> 40 #include <sys/pipe.h> 41 #include <sys/vnode.h> 42 #include <sys/uio.h> 43 #include <sys/event.h> 44 #include <sys/globaldata.h> 45 #include <sys/module.h> 46 #include <sys/malloc.h> 47 #include <sys/sysctl.h> 48 #include <sys/socket.h> 49 #include <sys/kern_syscall.h> 50 51 #include <vm/vm.h> 52 #include <vm/vm_param.h> 53 #include <sys/lock.h> 54 #include <vm/vm_object.h> 55 #include <vm/vm_kern.h> 56 #include <vm/vm_extern.h> 57 #include <vm/pmap.h> 58 #include <vm/vm_map.h> 59 #include <vm/vm_page.h> 60 #include <vm/vm_zone.h> 61 62 #include <sys/file2.h> 63 #include <sys/signal2.h> 64 65 #include <machine/cpufunc.h> 66 67 /* 68 * interfaces to the outside world 69 */ 70 static int pipe_read (struct file *fp, struct uio *uio, 71 struct ucred *cred, int flags); 72 static int pipe_write (struct file *fp, struct uio *uio, 73 struct ucred *cred, int flags); 74 static int pipe_close (struct file *fp); 75 static int pipe_shutdown (struct file *fp, int how); 76 static int pipe_kqfilter (struct file *fp, struct knote *kn); 77 static int pipe_stat (struct file *fp, struct stat *sb, struct ucred *cred); 78 static int pipe_ioctl (struct file *fp, u_long cmd, caddr_t data, 79 struct ucred *cred, struct sysmsg *msg); 80 81 static struct fileops pipeops = { 82 .fo_read = pipe_read, 83 .fo_write = pipe_write, 84 .fo_ioctl = pipe_ioctl, 85 .fo_kqfilter = pipe_kqfilter, 86 .fo_stat = pipe_stat, 87 .fo_close = pipe_close, 88 .fo_shutdown = pipe_shutdown 89 }; 90 91 static void filt_pipedetach(struct knote *kn); 92 static int filt_piperead(struct knote *kn, long hint); 93 static int filt_pipewrite(struct knote *kn, long hint); 94 95 static struct filterops pipe_rfiltops = 96 { FILTEROP_ISFD|FILTEROP_MPSAFE, NULL, filt_pipedetach, filt_piperead }; 97 static struct filterops pipe_wfiltops = 98 { FILTEROP_ISFD|FILTEROP_MPSAFE, NULL, filt_pipedetach, filt_pipewrite }; 99 100 MALLOC_DEFINE(M_PIPE, "pipe", "pipe structures"); 101 102 /* 103 * Default pipe buffer size(s), this can be kind-of large now because pipe 104 * space is pageable. The pipe code will try to maintain locality of 105 * reference for performance reasons, so small amounts of outstanding I/O 106 * will not wipe the cache. 107 */ 108 #define MINPIPESIZE (PIPE_SIZE/3) 109 #define MAXPIPESIZE (2*PIPE_SIZE/3) 110 111 /* 112 * Limit the number of "big" pipes 113 */ 114 #define LIMITBIGPIPES 64 115 #define PIPEQ_MAX_CACHE 16 /* per-cpu pipe structure cache */ 116 117 static int pipe_maxbig = LIMITBIGPIPES; 118 static int pipe_maxcache = PIPEQ_MAX_CACHE; 119 static int pipe_bigcount; 120 static int pipe_nbig; 121 static int pipe_bcache_alloc; 122 static int pipe_bkmem_alloc; 123 static int pipe_rblocked_count; 124 static int pipe_wblocked_count; 125 126 SYSCTL_NODE(_kern, OID_AUTO, pipe, CTLFLAG_RW, 0, "Pipe operation"); 127 SYSCTL_INT(_kern_pipe, OID_AUTO, nbig, 128 CTLFLAG_RD, &pipe_nbig, 0, "number of big pipes allocated"); 129 SYSCTL_INT(_kern_pipe, OID_AUTO, bigcount, 130 CTLFLAG_RW, &pipe_bigcount, 0, "number of times pipe expanded"); 131 SYSCTL_INT(_kern_pipe, OID_AUTO, rblocked, 132 CTLFLAG_RW, &pipe_rblocked_count, 0, "number of times pipe expanded"); 133 SYSCTL_INT(_kern_pipe, OID_AUTO, wblocked, 134 CTLFLAG_RW, &pipe_wblocked_count, 0, "number of times pipe expanded"); 135 SYSCTL_INT(_kern_pipe, OID_AUTO, maxcache, 136 CTLFLAG_RW, &pipe_maxcache, 0, "max pipes cached per-cpu"); 137 SYSCTL_INT(_kern_pipe, OID_AUTO, maxbig, 138 CTLFLAG_RW, &pipe_maxbig, 0, "max number of big pipes"); 139 static int pipe_delay = 5000; /* 5uS default */ 140 SYSCTL_INT(_kern_pipe, OID_AUTO, delay, 141 CTLFLAG_RW, &pipe_delay, 0, "SMP delay optimization in ns"); 142 #if !defined(NO_PIPE_SYSCTL_STATS) 143 SYSCTL_INT(_kern_pipe, OID_AUTO, bcache_alloc, 144 CTLFLAG_RW, &pipe_bcache_alloc, 0, "pipe buffer from pcpu cache"); 145 SYSCTL_INT(_kern_pipe, OID_AUTO, bkmem_alloc, 146 CTLFLAG_RW, &pipe_bkmem_alloc, 0, "pipe buffer from kmem"); 147 #endif 148 149 /* 150 * Auto-size pipe cache to reduce kmem allocations and frees. 151 */ 152 static 153 void 154 pipeinit(void *dummy) 155 { 156 size_t mbytes = kmem_lim_size(); 157 158 if (pipe_maxbig == LIMITBIGPIPES) { 159 if (mbytes >= 7 * 1024) 160 pipe_maxbig *= 2; 161 if (mbytes >= 15 * 1024) 162 pipe_maxbig *= 2; 163 } 164 if (pipe_maxcache == PIPEQ_MAX_CACHE) { 165 if (mbytes >= 7 * 1024) 166 pipe_maxcache *= 2; 167 if (mbytes >= 15 * 1024) 168 pipe_maxcache *= 2; 169 } 170 } 171 SYSINIT(kmem, SI_BOOT2_MACHDEP, SI_ORDER_ANY, pipeinit, NULL); 172 173 static void pipeclose (struct pipe *cpipe); 174 static void pipe_free_kmem (struct pipe *cpipe); 175 static int pipe_create (struct pipe **cpipep); 176 static int pipespace (struct pipe *cpipe, int size); 177 178 static __inline void 179 pipewakeup(struct pipe *cpipe, int dosigio) 180 { 181 if (dosigio && (cpipe->pipe_state & PIPE_ASYNC) && cpipe->pipe_sigio) { 182 lwkt_gettoken(&sigio_token); 183 pgsigio(cpipe->pipe_sigio, SIGIO, 0); 184 lwkt_reltoken(&sigio_token); 185 } 186 KNOTE(&cpipe->pipe_kq.ki_note, 0); 187 } 188 189 /* 190 * These routines are called before and after a UIO. The UIO 191 * may block, causing our held tokens to be lost temporarily. 192 * 193 * We use these routines to serialize reads against other reads 194 * and writes against other writes. 195 * 196 * The read token is held on entry so *ipp does not race. 197 */ 198 static __inline int 199 pipe_start_uio(struct pipe *cpipe, int *ipp) 200 { 201 int error; 202 203 while (*ipp) { 204 *ipp = -1; 205 error = tsleep(ipp, PCATCH, "pipexx", 0); 206 if (error) 207 return (error); 208 } 209 *ipp = 1; 210 return (0); 211 } 212 213 static __inline void 214 pipe_end_uio(struct pipe *cpipe, int *ipp) 215 { 216 if (*ipp < 0) { 217 *ipp = 0; 218 wakeup(ipp); 219 } else { 220 KKASSERT(*ipp > 0); 221 *ipp = 0; 222 } 223 } 224 225 /* 226 * The pipe system call for the DTYPE_PIPE type of pipes 227 * 228 * pipe_args(int dummy) 229 * 230 * MPSAFE 231 */ 232 int 233 sys_pipe(struct pipe_args *uap) 234 { 235 return kern_pipe(uap->sysmsg_fds, 0); 236 } 237 238 int 239 sys_pipe2(struct pipe2_args *uap) 240 { 241 return kern_pipe(uap->sysmsg_fds, uap->flags); 242 } 243 244 int 245 kern_pipe(long *fds, int flags) 246 { 247 struct thread *td = curthread; 248 struct filedesc *fdp = td->td_proc->p_fd; 249 struct file *rf, *wf; 250 struct pipe *rpipe, *wpipe; 251 int fd1, fd2, error; 252 253 rpipe = wpipe = NULL; 254 if (pipe_create(&rpipe) || pipe_create(&wpipe)) { 255 pipeclose(rpipe); 256 pipeclose(wpipe); 257 return (ENFILE); 258 } 259 260 error = falloc(td->td_lwp, &rf, &fd1); 261 if (error) { 262 pipeclose(rpipe); 263 pipeclose(wpipe); 264 return (error); 265 } 266 fds[0] = fd1; 267 268 /* 269 * Warning: once we've gotten past allocation of the fd for the 270 * read-side, we can only drop the read side via fdrop() in order 271 * to avoid races against processes which manage to dup() the read 272 * side while we are blocked trying to allocate the write side. 273 */ 274 rf->f_type = DTYPE_PIPE; 275 rf->f_flag = FREAD | FWRITE; 276 rf->f_ops = &pipeops; 277 rf->f_data = rpipe; 278 if (flags & O_NONBLOCK) 279 rf->f_flag |= O_NONBLOCK; 280 if (flags & O_CLOEXEC) 281 fdp->fd_files[fd1].fileflags |= UF_EXCLOSE; 282 283 error = falloc(td->td_lwp, &wf, &fd2); 284 if (error) { 285 fsetfd(fdp, NULL, fd1); 286 fdrop(rf); 287 /* rpipe has been closed by fdrop(). */ 288 pipeclose(wpipe); 289 return (error); 290 } 291 wf->f_type = DTYPE_PIPE; 292 wf->f_flag = FREAD | FWRITE; 293 wf->f_ops = &pipeops; 294 wf->f_data = wpipe; 295 if (flags & O_NONBLOCK) 296 wf->f_flag |= O_NONBLOCK; 297 if (flags & O_CLOEXEC) 298 fdp->fd_files[fd2].fileflags |= UF_EXCLOSE; 299 300 fds[1] = fd2; 301 302 rpipe->pipe_slock = kmalloc(sizeof(struct lock), 303 M_PIPE, M_WAITOK|M_ZERO); 304 wpipe->pipe_slock = rpipe->pipe_slock; 305 rpipe->pipe_peer = wpipe; 306 wpipe->pipe_peer = rpipe; 307 lockinit(rpipe->pipe_slock, "pipecl", 0, 0); 308 309 /* 310 * Once activated the peer relationship remains valid until 311 * both sides are closed. 312 */ 313 fsetfd(fdp, rf, fd1); 314 fsetfd(fdp, wf, fd2); 315 fdrop(rf); 316 fdrop(wf); 317 318 return (0); 319 } 320 321 /* 322 * Allocate kva for pipe circular buffer, the space is pageable 323 * This routine will 'realloc' the size of a pipe safely, if it fails 324 * it will retain the old buffer. 325 * If it fails it will return ENOMEM. 326 */ 327 static int 328 pipespace(struct pipe *cpipe, int size) 329 { 330 struct vm_object *object; 331 caddr_t buffer; 332 int npages, error; 333 334 npages = round_page(size) / PAGE_SIZE; 335 object = cpipe->pipe_buffer.object; 336 337 /* 338 * [re]create the object if necessary and reserve space for it 339 * in the kernel_map. The object and memory are pageable. On 340 * success, free the old resources before assigning the new 341 * ones. 342 */ 343 if (object == NULL || object->size != npages) { 344 object = vm_object_allocate(OBJT_DEFAULT, npages); 345 buffer = (caddr_t)vm_map_min(&kernel_map); 346 347 error = vm_map_find(&kernel_map, object, NULL, 348 0, (vm_offset_t *)&buffer, size, 349 PAGE_SIZE, 350 1, VM_MAPTYPE_NORMAL, 351 VM_PROT_ALL, VM_PROT_ALL, 0); 352 353 if (error != KERN_SUCCESS) { 354 vm_object_deallocate(object); 355 return (ENOMEM); 356 } 357 pipe_free_kmem(cpipe); 358 cpipe->pipe_buffer.object = object; 359 cpipe->pipe_buffer.buffer = buffer; 360 cpipe->pipe_buffer.size = size; 361 ++pipe_bkmem_alloc; 362 } else { 363 ++pipe_bcache_alloc; 364 } 365 cpipe->pipe_buffer.rindex = 0; 366 cpipe->pipe_buffer.windex = 0; 367 return (0); 368 } 369 370 /* 371 * Initialize and allocate VM and memory for pipe, pulling the pipe from 372 * our per-cpu cache if possible. For now make sure it is sized for the 373 * smaller PIPE_SIZE default. 374 */ 375 static int 376 pipe_create(struct pipe **cpipep) 377 { 378 globaldata_t gd = mycpu; 379 struct pipe *cpipe; 380 int error; 381 382 if ((cpipe = gd->gd_pipeq) != NULL) { 383 gd->gd_pipeq = cpipe->pipe_peer; 384 --gd->gd_pipeqcount; 385 cpipe->pipe_peer = NULL; 386 cpipe->pipe_wantwcnt = 0; 387 } else { 388 cpipe = kmalloc(sizeof(struct pipe), M_PIPE, M_WAITOK|M_ZERO); 389 } 390 *cpipep = cpipe; 391 if ((error = pipespace(cpipe, PIPE_SIZE)) != 0) 392 return (error); 393 vfs_timestamp(&cpipe->pipe_ctime); 394 cpipe->pipe_atime = cpipe->pipe_ctime; 395 cpipe->pipe_mtime = cpipe->pipe_ctime; 396 lwkt_token_init(&cpipe->pipe_rlock, "piper"); 397 lwkt_token_init(&cpipe->pipe_wlock, "pipew"); 398 return (0); 399 } 400 401 static int 402 pipe_read(struct file *fp, struct uio *uio, struct ucred *cred, int fflags) 403 { 404 struct pipe *rpipe; 405 struct pipe *wpipe; 406 int error; 407 size_t nread = 0; 408 int nbio; 409 u_int size; /* total bytes available */ 410 u_int nsize; /* total bytes to read */ 411 u_int rindex; /* contiguous bytes available */ 412 int notify_writer; 413 int bigread; 414 int bigcount; 415 416 atomic_set_int(&curthread->td_mpflags, TDF_MP_BATCH_DEMARC); 417 418 if (uio->uio_resid == 0) 419 return(0); 420 421 /* 422 * Setup locks, calculate nbio 423 */ 424 rpipe = (struct pipe *)fp->f_data; 425 wpipe = rpipe->pipe_peer; 426 lwkt_gettoken(&rpipe->pipe_rlock); 427 428 if (fflags & O_FBLOCKING) 429 nbio = 0; 430 else if (fflags & O_FNONBLOCKING) 431 nbio = 1; 432 else if (fp->f_flag & O_NONBLOCK) 433 nbio = 1; 434 else 435 nbio = 0; 436 437 /* 438 * Reads are serialized. Note however that pipe_buffer.buffer and 439 * pipe_buffer.size can change out from under us when the number 440 * of bytes in the buffer are zero due to the write-side doing a 441 * pipespace(). 442 */ 443 error = pipe_start_uio(rpipe, &rpipe->pipe_rip); 444 if (error) { 445 lwkt_reltoken(&rpipe->pipe_rlock); 446 return (error); 447 } 448 notify_writer = 0; 449 450 bigread = (uio->uio_resid > 10 * 1024 * 1024); 451 bigcount = 10; 452 453 while (uio->uio_resid) { 454 /* 455 * Don't hog the cpu. 456 */ 457 if (bigread && --bigcount == 0) { 458 lwkt_user_yield(); 459 bigcount = 10; 460 if (CURSIG(curthread->td_lwp)) { 461 error = EINTR; 462 break; 463 } 464 } 465 466 size = rpipe->pipe_buffer.windex - rpipe->pipe_buffer.rindex; 467 cpu_lfence(); 468 if (size) { 469 rindex = rpipe->pipe_buffer.rindex & 470 (rpipe->pipe_buffer.size - 1); 471 nsize = size; 472 if (nsize > rpipe->pipe_buffer.size - rindex) 473 nsize = rpipe->pipe_buffer.size - rindex; 474 nsize = szmin(nsize, uio->uio_resid); 475 476 error = uiomove(&rpipe->pipe_buffer.buffer[rindex], 477 nsize, uio); 478 if (error) 479 break; 480 cpu_mfence(); 481 rpipe->pipe_buffer.rindex += nsize; 482 nread += nsize; 483 484 /* 485 * If the FIFO is still over half full just continue 486 * and do not try to notify the writer yet. 487 */ 488 if (size - nsize >= (rpipe->pipe_buffer.size >> 1)) { 489 notify_writer = 0; 490 continue; 491 } 492 493 /* 494 * When the FIFO is less then half full notify any 495 * waiting writer. WANTW can be checked while 496 * holding just the rlock. 497 */ 498 notify_writer = 1; 499 if ((rpipe->pipe_state & PIPE_WANTW) == 0) 500 continue; 501 } 502 503 /* 504 * If the "write-side" was blocked we wake it up. This code 505 * is reached either when the buffer is completely emptied 506 * or if it becomes more then half-empty. 507 * 508 * Pipe_state can only be modified if both the rlock and 509 * wlock are held. 510 */ 511 if (rpipe->pipe_state & PIPE_WANTW) { 512 lwkt_gettoken(&rpipe->pipe_wlock); 513 if (rpipe->pipe_state & PIPE_WANTW) { 514 rpipe->pipe_state &= ~PIPE_WANTW; 515 lwkt_reltoken(&rpipe->pipe_wlock); 516 wakeup(rpipe); 517 } else { 518 lwkt_reltoken(&rpipe->pipe_wlock); 519 } 520 } 521 522 /* 523 * Pick up our copy loop again if the writer sent data to 524 * us while we were messing around. 525 * 526 * On a SMP box poll up to pipe_delay nanoseconds for new 527 * data. Typically a value of 2000 to 4000 is sufficient 528 * to eradicate most IPIs/tsleeps/wakeups when a pipe 529 * is used for synchronous communications with small packets, 530 * and 8000 or so (8uS) will pipeline large buffer xfers 531 * between cpus over a pipe. 532 * 533 * For synchronous communications a hit means doing a 534 * full Awrite-Bread-Bwrite-Aread cycle in less then 2uS, 535 * where as miss requiring a tsleep/wakeup sequence 536 * will take 7uS or more. 537 */ 538 if (rpipe->pipe_buffer.windex != rpipe->pipe_buffer.rindex) 539 continue; 540 541 #ifdef _RDTSC_SUPPORTED_ 542 if (pipe_delay) { 543 int64_t tsc_target; 544 int good = 0; 545 546 tsc_target = tsc_get_target(pipe_delay); 547 while (tsc_test_target(tsc_target) == 0) { 548 if (rpipe->pipe_buffer.windex != 549 rpipe->pipe_buffer.rindex) { 550 good = 1; 551 break; 552 } 553 } 554 if (good) 555 continue; 556 } 557 #endif 558 559 /* 560 * Detect EOF condition, do not set error. 561 */ 562 if (rpipe->pipe_state & PIPE_REOF) 563 break; 564 565 /* 566 * Break if some data was read, or if this was a non-blocking 567 * read. 568 */ 569 if (nread > 0) 570 break; 571 572 if (nbio) { 573 error = EAGAIN; 574 break; 575 } 576 577 /* 578 * Last chance, interlock with WANTR. 579 */ 580 lwkt_gettoken(&rpipe->pipe_wlock); 581 size = rpipe->pipe_buffer.windex - rpipe->pipe_buffer.rindex; 582 if (size) { 583 lwkt_reltoken(&rpipe->pipe_wlock); 584 continue; 585 } 586 587 /* 588 * Retest EOF - acquiring a new token can temporarily release 589 * tokens already held. 590 */ 591 if (rpipe->pipe_state & PIPE_REOF) { 592 lwkt_reltoken(&rpipe->pipe_wlock); 593 break; 594 } 595 596 /* 597 * If there is no more to read in the pipe, reset its 598 * pointers to the beginning. This improves cache hit 599 * stats. 600 * 601 * We need both locks to modify both pointers, and there 602 * must also not be a write in progress or the uiomove() 603 * in the write might block and temporarily release 604 * its wlock, then reacquire and update windex. We are 605 * only serialized against reads, not writes. 606 * 607 * XXX should we even bother resetting the indices? It 608 * might actually be more cache efficient not to. 609 */ 610 if (rpipe->pipe_buffer.rindex == rpipe->pipe_buffer.windex && 611 rpipe->pipe_wip == 0) { 612 rpipe->pipe_buffer.rindex = 0; 613 rpipe->pipe_buffer.windex = 0; 614 } 615 616 /* 617 * Wait for more data. 618 * 619 * Pipe_state can only be set if both the rlock and wlock 620 * are held. 621 */ 622 rpipe->pipe_state |= PIPE_WANTR; 623 tsleep_interlock(rpipe, PCATCH); 624 lwkt_reltoken(&rpipe->pipe_wlock); 625 error = tsleep(rpipe, PCATCH | PINTERLOCKED, "piperd", 0); 626 ++pipe_rblocked_count; 627 if (error) 628 break; 629 } 630 pipe_end_uio(rpipe, &rpipe->pipe_rip); 631 632 /* 633 * Uptime last access time 634 */ 635 if (error == 0 && nread) 636 vfs_timestamp(&rpipe->pipe_atime); 637 638 /* 639 * If we drained the FIFO more then half way then handle 640 * write blocking hysteresis. 641 * 642 * Note that PIPE_WANTW cannot be set by the writer without 643 * it holding both rlock and wlock, so we can test it 644 * while holding just rlock. 645 */ 646 if (notify_writer) { 647 /* 648 * Synchronous blocking is done on the pipe involved 649 */ 650 if (rpipe->pipe_state & PIPE_WANTW) { 651 lwkt_gettoken(&rpipe->pipe_wlock); 652 if (rpipe->pipe_state & PIPE_WANTW) { 653 rpipe->pipe_state &= ~PIPE_WANTW; 654 lwkt_reltoken(&rpipe->pipe_wlock); 655 wakeup(rpipe); 656 } else { 657 lwkt_reltoken(&rpipe->pipe_wlock); 658 } 659 } 660 661 /* 662 * But we may also have to deal with a kqueue which is 663 * stored on the same pipe as its descriptor, so a 664 * EVFILT_WRITE event waiting for our side to drain will 665 * be on the other side. 666 */ 667 lwkt_gettoken(&wpipe->pipe_wlock); 668 pipewakeup(wpipe, 0); 669 lwkt_reltoken(&wpipe->pipe_wlock); 670 } 671 /*size = rpipe->pipe_buffer.windex - rpipe->pipe_buffer.rindex;*/ 672 lwkt_reltoken(&rpipe->pipe_rlock); 673 674 return (error); 675 } 676 677 static int 678 pipe_write(struct file *fp, struct uio *uio, struct ucred *cred, int fflags) 679 { 680 int error; 681 int orig_resid; 682 int nbio; 683 struct pipe *wpipe; 684 struct pipe *rpipe; 685 u_int windex; 686 u_int space; 687 u_int wcount; 688 int bigwrite; 689 int bigcount; 690 691 /* 692 * Writes go to the peer. The peer will always exist. 693 */ 694 rpipe = (struct pipe *) fp->f_data; 695 wpipe = rpipe->pipe_peer; 696 lwkt_gettoken(&wpipe->pipe_wlock); 697 if (wpipe->pipe_state & PIPE_WEOF) { 698 lwkt_reltoken(&wpipe->pipe_wlock); 699 return (EPIPE); 700 } 701 702 /* 703 * Degenerate case (EPIPE takes prec) 704 */ 705 if (uio->uio_resid == 0) { 706 lwkt_reltoken(&wpipe->pipe_wlock); 707 return(0); 708 } 709 710 /* 711 * Writes are serialized (start_uio must be called with wlock) 712 */ 713 error = pipe_start_uio(wpipe, &wpipe->pipe_wip); 714 if (error) { 715 lwkt_reltoken(&wpipe->pipe_wlock); 716 return (error); 717 } 718 719 if (fflags & O_FBLOCKING) 720 nbio = 0; 721 else if (fflags & O_FNONBLOCKING) 722 nbio = 1; 723 else if (fp->f_flag & O_NONBLOCK) 724 nbio = 1; 725 else 726 nbio = 0; 727 728 /* 729 * If it is advantageous to resize the pipe buffer, do 730 * so. We are write-serialized so we can block safely. 731 */ 732 if ((wpipe->pipe_buffer.size <= PIPE_SIZE) && 733 (pipe_nbig < pipe_maxbig) && 734 wpipe->pipe_wantwcnt > 4 && 735 (wpipe->pipe_buffer.rindex == wpipe->pipe_buffer.windex)) { 736 /* 737 * Recheck after lock. 738 */ 739 lwkt_gettoken(&wpipe->pipe_rlock); 740 if ((wpipe->pipe_buffer.size <= PIPE_SIZE) && 741 (pipe_nbig < pipe_maxbig) && 742 (wpipe->pipe_buffer.rindex == wpipe->pipe_buffer.windex)) { 743 atomic_add_int(&pipe_nbig, 1); 744 if (pipespace(wpipe, BIG_PIPE_SIZE) == 0) 745 ++pipe_bigcount; 746 else 747 atomic_subtract_int(&pipe_nbig, 1); 748 } 749 lwkt_reltoken(&wpipe->pipe_rlock); 750 } 751 752 orig_resid = uio->uio_resid; 753 wcount = 0; 754 755 bigwrite = (uio->uio_resid > 10 * 1024 * 1024); 756 bigcount = 10; 757 758 while (uio->uio_resid) { 759 if (wpipe->pipe_state & PIPE_WEOF) { 760 error = EPIPE; 761 break; 762 } 763 764 /* 765 * Don't hog the cpu. 766 */ 767 if (bigwrite && --bigcount == 0) { 768 lwkt_user_yield(); 769 bigcount = 10; 770 if (CURSIG(curthread->td_lwp)) { 771 error = EINTR; 772 break; 773 } 774 } 775 776 windex = wpipe->pipe_buffer.windex & 777 (wpipe->pipe_buffer.size - 1); 778 space = wpipe->pipe_buffer.size - 779 (wpipe->pipe_buffer.windex - wpipe->pipe_buffer.rindex); 780 cpu_lfence(); 781 782 /* Writes of size <= PIPE_BUF must be atomic. */ 783 if ((space < uio->uio_resid) && (orig_resid <= PIPE_BUF)) 784 space = 0; 785 786 /* 787 * Write to fill, read size handles write hysteresis. Also 788 * additional restrictions can cause select-based non-blocking 789 * writes to spin. 790 */ 791 if (space > 0) { 792 u_int segsize; 793 794 /* 795 * Transfer size is minimum of uio transfer 796 * and free space in pipe buffer. 797 * 798 * Limit each uiocopy to no more then PIPE_SIZE 799 * so we can keep the gravy train going on a 800 * SMP box. This doubles the performance for 801 * write sizes > 16K. Otherwise large writes 802 * wind up doing an inefficient synchronous 803 * ping-pong. 804 */ 805 space = szmin(space, uio->uio_resid); 806 if (space > PIPE_SIZE) 807 space = PIPE_SIZE; 808 809 /* 810 * First segment to transfer is minimum of 811 * transfer size and contiguous space in 812 * pipe buffer. If first segment to transfer 813 * is less than the transfer size, we've got 814 * a wraparound in the buffer. 815 */ 816 segsize = wpipe->pipe_buffer.size - windex; 817 if (segsize > space) 818 segsize = space; 819 820 /* 821 * If this is the first loop and the reader is 822 * blocked, do a preemptive wakeup of the reader. 823 * 824 * On SMP the IPI latency plus the wlock interlock 825 * on the reader side is the fastest way to get the 826 * reader going. (The scheduler will hard loop on 827 * lock tokens). 828 * 829 * NOTE: We can't clear WANTR here without acquiring 830 * the rlock, which we don't want to do here! 831 */ 832 if ((wpipe->pipe_state & PIPE_WANTR)) 833 wakeup(wpipe); 834 835 /* 836 * Transfer segment, which may include a wrap-around. 837 * Update windex to account for both all in one go 838 * so the reader can read() the data atomically. 839 */ 840 error = uiomove(&wpipe->pipe_buffer.buffer[windex], 841 segsize, uio); 842 if (error == 0 && segsize < space) { 843 segsize = space - segsize; 844 error = uiomove(&wpipe->pipe_buffer.buffer[0], 845 segsize, uio); 846 } 847 if (error) 848 break; 849 cpu_mfence(); 850 wpipe->pipe_buffer.windex += space; 851 wcount += space; 852 continue; 853 } 854 855 /* 856 * We need both the rlock and the wlock to interlock against 857 * the EOF, WANTW, and size checks, and to modify pipe_state. 858 * 859 * These are token locks so we do not have to worry about 860 * deadlocks. 861 */ 862 lwkt_gettoken(&wpipe->pipe_rlock); 863 864 /* 865 * If the "read-side" has been blocked, wake it up now 866 * and yield to let it drain synchronously rather 867 * then block. 868 */ 869 if (wpipe->pipe_state & PIPE_WANTR) { 870 wpipe->pipe_state &= ~PIPE_WANTR; 871 wakeup(wpipe); 872 } 873 874 /* 875 * don't block on non-blocking I/O 876 */ 877 if (nbio) { 878 lwkt_reltoken(&wpipe->pipe_rlock); 879 error = EAGAIN; 880 break; 881 } 882 883 /* 884 * re-test whether we have to block in the writer after 885 * acquiring both locks, in case the reader opened up 886 * some space. 887 */ 888 space = wpipe->pipe_buffer.size - 889 (wpipe->pipe_buffer.windex - wpipe->pipe_buffer.rindex); 890 cpu_lfence(); 891 if ((space < uio->uio_resid) && (orig_resid <= PIPE_BUF)) 892 space = 0; 893 894 /* 895 * Retest EOF - acquiring a new token can temporarily release 896 * tokens already held. 897 */ 898 if (wpipe->pipe_state & PIPE_WEOF) { 899 lwkt_reltoken(&wpipe->pipe_rlock); 900 error = EPIPE; 901 break; 902 } 903 904 /* 905 * We have no more space and have something to offer, 906 * wake up select/poll/kq. 907 */ 908 if (space == 0) { 909 wpipe->pipe_state |= PIPE_WANTW; 910 ++wpipe->pipe_wantwcnt; 911 pipewakeup(wpipe, 1); 912 if (wpipe->pipe_state & PIPE_WANTW) 913 error = tsleep(wpipe, PCATCH, "pipewr", 0); 914 ++pipe_wblocked_count; 915 } 916 lwkt_reltoken(&wpipe->pipe_rlock); 917 918 /* 919 * Break out if we errored or the read side wants us to go 920 * away. 921 */ 922 if (error) 923 break; 924 if (wpipe->pipe_state & PIPE_WEOF) { 925 error = EPIPE; 926 break; 927 } 928 } 929 pipe_end_uio(wpipe, &wpipe->pipe_wip); 930 931 /* 932 * If we have put any characters in the buffer, we wake up 933 * the reader. 934 * 935 * Both rlock and wlock are required to be able to modify pipe_state. 936 */ 937 if (wpipe->pipe_buffer.windex != wpipe->pipe_buffer.rindex) { 938 if (wpipe->pipe_state & PIPE_WANTR) { 939 lwkt_gettoken(&wpipe->pipe_rlock); 940 if (wpipe->pipe_state & PIPE_WANTR) { 941 wpipe->pipe_state &= ~PIPE_WANTR; 942 lwkt_reltoken(&wpipe->pipe_rlock); 943 wakeup(wpipe); 944 } else { 945 lwkt_reltoken(&wpipe->pipe_rlock); 946 } 947 } 948 lwkt_gettoken(&wpipe->pipe_rlock); 949 pipewakeup(wpipe, 1); 950 lwkt_reltoken(&wpipe->pipe_rlock); 951 } 952 953 /* 954 * Don't return EPIPE if I/O was successful 955 */ 956 if ((wpipe->pipe_buffer.rindex == wpipe->pipe_buffer.windex) && 957 (uio->uio_resid == 0) && 958 (error == EPIPE)) { 959 error = 0; 960 } 961 962 if (error == 0) 963 vfs_timestamp(&wpipe->pipe_mtime); 964 965 /* 966 * We have something to offer, 967 * wake up select/poll/kq. 968 */ 969 /*space = wpipe->pipe_buffer.windex - wpipe->pipe_buffer.rindex;*/ 970 lwkt_reltoken(&wpipe->pipe_wlock); 971 return (error); 972 } 973 974 /* 975 * we implement a very minimal set of ioctls for compatibility with sockets. 976 */ 977 static int 978 pipe_ioctl(struct file *fp, u_long cmd, caddr_t data, 979 struct ucred *cred, struct sysmsg *msg) 980 { 981 struct pipe *mpipe; 982 int error; 983 984 mpipe = (struct pipe *)fp->f_data; 985 986 lwkt_gettoken(&mpipe->pipe_rlock); 987 lwkt_gettoken(&mpipe->pipe_wlock); 988 989 switch (cmd) { 990 case FIOASYNC: 991 if (*(int *)data) { 992 mpipe->pipe_state |= PIPE_ASYNC; 993 } else { 994 mpipe->pipe_state &= ~PIPE_ASYNC; 995 } 996 error = 0; 997 break; 998 case FIONREAD: 999 *(int *)data = mpipe->pipe_buffer.windex - 1000 mpipe->pipe_buffer.rindex; 1001 error = 0; 1002 break; 1003 case FIOSETOWN: 1004 error = fsetown(*(int *)data, &mpipe->pipe_sigio); 1005 break; 1006 case FIOGETOWN: 1007 *(int *)data = fgetown(&mpipe->pipe_sigio); 1008 error = 0; 1009 break; 1010 case TIOCSPGRP: 1011 /* This is deprecated, FIOSETOWN should be used instead. */ 1012 error = fsetown(-(*(int *)data), &mpipe->pipe_sigio); 1013 break; 1014 1015 case TIOCGPGRP: 1016 /* This is deprecated, FIOGETOWN should be used instead. */ 1017 *(int *)data = -fgetown(&mpipe->pipe_sigio); 1018 error = 0; 1019 break; 1020 default: 1021 error = ENOTTY; 1022 break; 1023 } 1024 lwkt_reltoken(&mpipe->pipe_wlock); 1025 lwkt_reltoken(&mpipe->pipe_rlock); 1026 1027 return (error); 1028 } 1029 1030 /* 1031 * MPSAFE 1032 */ 1033 static int 1034 pipe_stat(struct file *fp, struct stat *ub, struct ucred *cred) 1035 { 1036 struct pipe *pipe; 1037 1038 pipe = (struct pipe *)fp->f_data; 1039 1040 bzero((caddr_t)ub, sizeof(*ub)); 1041 ub->st_mode = S_IFIFO; 1042 ub->st_blksize = pipe->pipe_buffer.size; 1043 ub->st_size = pipe->pipe_buffer.windex - pipe->pipe_buffer.rindex; 1044 ub->st_blocks = (ub->st_size + ub->st_blksize - 1) / ub->st_blksize; 1045 ub->st_atimespec = pipe->pipe_atime; 1046 ub->st_mtimespec = pipe->pipe_mtime; 1047 ub->st_ctimespec = pipe->pipe_ctime; 1048 /* 1049 * Left as 0: st_dev, st_ino, st_nlink, st_uid, st_gid, st_rdev, 1050 * st_flags, st_gen. 1051 * XXX (st_dev, st_ino) should be unique. 1052 */ 1053 return (0); 1054 } 1055 1056 static int 1057 pipe_close(struct file *fp) 1058 { 1059 struct pipe *cpipe; 1060 1061 cpipe = (struct pipe *)fp->f_data; 1062 fp->f_ops = &badfileops; 1063 fp->f_data = NULL; 1064 funsetown(&cpipe->pipe_sigio); 1065 pipeclose(cpipe); 1066 return (0); 1067 } 1068 1069 /* 1070 * Shutdown one or both directions of a full-duplex pipe. 1071 */ 1072 static int 1073 pipe_shutdown(struct file *fp, int how) 1074 { 1075 struct pipe *rpipe; 1076 struct pipe *wpipe; 1077 int error = EPIPE; 1078 1079 rpipe = (struct pipe *)fp->f_data; 1080 wpipe = rpipe->pipe_peer; 1081 1082 /* 1083 * We modify pipe_state on both pipes, which means we need 1084 * all four tokens! 1085 */ 1086 lwkt_gettoken(&rpipe->pipe_rlock); 1087 lwkt_gettoken(&rpipe->pipe_wlock); 1088 lwkt_gettoken(&wpipe->pipe_rlock); 1089 lwkt_gettoken(&wpipe->pipe_wlock); 1090 1091 switch(how) { 1092 case SHUT_RDWR: 1093 case SHUT_RD: 1094 rpipe->pipe_state |= PIPE_REOF; /* my reads */ 1095 rpipe->pipe_state |= PIPE_WEOF; /* peer writes */ 1096 if (rpipe->pipe_state & PIPE_WANTR) { 1097 rpipe->pipe_state &= ~PIPE_WANTR; 1098 wakeup(rpipe); 1099 } 1100 if (rpipe->pipe_state & PIPE_WANTW) { 1101 rpipe->pipe_state &= ~PIPE_WANTW; 1102 wakeup(rpipe); 1103 } 1104 error = 0; 1105 if (how == SHUT_RD) 1106 break; 1107 /* fall through */ 1108 case SHUT_WR: 1109 wpipe->pipe_state |= PIPE_REOF; /* peer reads */ 1110 wpipe->pipe_state |= PIPE_WEOF; /* my writes */ 1111 if (wpipe->pipe_state & PIPE_WANTR) { 1112 wpipe->pipe_state &= ~PIPE_WANTR; 1113 wakeup(wpipe); 1114 } 1115 if (wpipe->pipe_state & PIPE_WANTW) { 1116 wpipe->pipe_state &= ~PIPE_WANTW; 1117 wakeup(wpipe); 1118 } 1119 error = 0; 1120 break; 1121 } 1122 pipewakeup(rpipe, 1); 1123 pipewakeup(wpipe, 1); 1124 1125 lwkt_reltoken(&wpipe->pipe_wlock); 1126 lwkt_reltoken(&wpipe->pipe_rlock); 1127 lwkt_reltoken(&rpipe->pipe_wlock); 1128 lwkt_reltoken(&rpipe->pipe_rlock); 1129 1130 return (error); 1131 } 1132 1133 static void 1134 pipe_free_kmem(struct pipe *cpipe) 1135 { 1136 if (cpipe->pipe_buffer.buffer != NULL) { 1137 if (cpipe->pipe_buffer.size > PIPE_SIZE) 1138 atomic_subtract_int(&pipe_nbig, 1); 1139 kmem_free(&kernel_map, 1140 (vm_offset_t)cpipe->pipe_buffer.buffer, 1141 cpipe->pipe_buffer.size); 1142 cpipe->pipe_buffer.buffer = NULL; 1143 cpipe->pipe_buffer.object = NULL; 1144 } 1145 } 1146 1147 /* 1148 * Close the pipe. The slock must be held to interlock against simultanious 1149 * closes. The rlock and wlock must be held to adjust the pipe_state. 1150 */ 1151 static void 1152 pipeclose(struct pipe *cpipe) 1153 { 1154 globaldata_t gd; 1155 struct pipe *ppipe; 1156 1157 if (cpipe == NULL) 1158 return; 1159 1160 /* 1161 * The slock may not have been allocated yet (close during 1162 * initialization) 1163 * 1164 * We need both the read and write tokens to modify pipe_state. 1165 */ 1166 if (cpipe->pipe_slock) 1167 lockmgr(cpipe->pipe_slock, LK_EXCLUSIVE); 1168 lwkt_gettoken(&cpipe->pipe_rlock); 1169 lwkt_gettoken(&cpipe->pipe_wlock); 1170 1171 /* 1172 * Set our state, wakeup anyone waiting in select/poll/kq, and 1173 * wakeup anyone blocked on our pipe. 1174 */ 1175 cpipe->pipe_state |= PIPE_CLOSED | PIPE_REOF | PIPE_WEOF; 1176 pipewakeup(cpipe, 1); 1177 if (cpipe->pipe_state & (PIPE_WANTR | PIPE_WANTW)) { 1178 cpipe->pipe_state &= ~(PIPE_WANTR | PIPE_WANTW); 1179 wakeup(cpipe); 1180 } 1181 1182 /* 1183 * Disconnect from peer. 1184 */ 1185 if ((ppipe = cpipe->pipe_peer) != NULL) { 1186 lwkt_gettoken(&ppipe->pipe_rlock); 1187 lwkt_gettoken(&ppipe->pipe_wlock); 1188 ppipe->pipe_state |= PIPE_REOF | PIPE_WEOF; 1189 pipewakeup(ppipe, 1); 1190 if (ppipe->pipe_state & (PIPE_WANTR | PIPE_WANTW)) { 1191 ppipe->pipe_state &= ~(PIPE_WANTR | PIPE_WANTW); 1192 wakeup(ppipe); 1193 } 1194 if (SLIST_FIRST(&ppipe->pipe_kq.ki_note)) 1195 KNOTE(&ppipe->pipe_kq.ki_note, 0); 1196 lwkt_reltoken(&ppipe->pipe_wlock); 1197 lwkt_reltoken(&ppipe->pipe_rlock); 1198 } 1199 1200 /* 1201 * If the peer is also closed we can free resources for both 1202 * sides, otherwise we leave our side intact to deal with any 1203 * races (since we only have the slock). 1204 */ 1205 if (ppipe && (ppipe->pipe_state & PIPE_CLOSED)) { 1206 cpipe->pipe_peer = NULL; 1207 ppipe->pipe_peer = NULL; 1208 ppipe->pipe_slock = NULL; /* we will free the slock */ 1209 pipeclose(ppipe); 1210 ppipe = NULL; 1211 } 1212 1213 lwkt_reltoken(&cpipe->pipe_wlock); 1214 lwkt_reltoken(&cpipe->pipe_rlock); 1215 if (cpipe->pipe_slock) 1216 lockmgr(cpipe->pipe_slock, LK_RELEASE); 1217 1218 /* 1219 * If we disassociated from our peer we can free resources 1220 */ 1221 if (ppipe == NULL) { 1222 gd = mycpu; 1223 if (cpipe->pipe_slock) { 1224 kfree(cpipe->pipe_slock, M_PIPE); 1225 cpipe->pipe_slock = NULL; 1226 } 1227 if (gd->gd_pipeqcount >= pipe_maxcache || 1228 cpipe->pipe_buffer.size != PIPE_SIZE 1229 ) { 1230 pipe_free_kmem(cpipe); 1231 kfree(cpipe, M_PIPE); 1232 } else { 1233 cpipe->pipe_state = 0; 1234 cpipe->pipe_peer = gd->gd_pipeq; 1235 gd->gd_pipeq = cpipe; 1236 ++gd->gd_pipeqcount; 1237 } 1238 } 1239 } 1240 1241 static int 1242 pipe_kqfilter(struct file *fp, struct knote *kn) 1243 { 1244 struct pipe *cpipe; 1245 1246 cpipe = (struct pipe *)kn->kn_fp->f_data; 1247 1248 switch (kn->kn_filter) { 1249 case EVFILT_READ: 1250 kn->kn_fop = &pipe_rfiltops; 1251 break; 1252 case EVFILT_WRITE: 1253 kn->kn_fop = &pipe_wfiltops; 1254 if (cpipe->pipe_peer == NULL) { 1255 /* other end of pipe has been closed */ 1256 return (EPIPE); 1257 } 1258 break; 1259 default: 1260 return (EOPNOTSUPP); 1261 } 1262 kn->kn_hook = (caddr_t)cpipe; 1263 1264 knote_insert(&cpipe->pipe_kq.ki_note, kn); 1265 1266 return (0); 1267 } 1268 1269 static void 1270 filt_pipedetach(struct knote *kn) 1271 { 1272 struct pipe *cpipe = (struct pipe *)kn->kn_hook; 1273 1274 knote_remove(&cpipe->pipe_kq.ki_note, kn); 1275 } 1276 1277 /*ARGSUSED*/ 1278 static int 1279 filt_piperead(struct knote *kn, long hint) 1280 { 1281 struct pipe *rpipe = (struct pipe *)kn->kn_fp->f_data; 1282 int ready = 0; 1283 1284 lwkt_gettoken(&rpipe->pipe_rlock); 1285 lwkt_gettoken(&rpipe->pipe_wlock); 1286 1287 kn->kn_data = rpipe->pipe_buffer.windex - rpipe->pipe_buffer.rindex; 1288 1289 if (rpipe->pipe_state & PIPE_REOF) { 1290 /* 1291 * Only set NODATA if all data has been exhausted 1292 */ 1293 if (kn->kn_data == 0) 1294 kn->kn_flags |= EV_NODATA; 1295 kn->kn_flags |= EV_EOF; 1296 ready = 1; 1297 } 1298 1299 lwkt_reltoken(&rpipe->pipe_wlock); 1300 lwkt_reltoken(&rpipe->pipe_rlock); 1301 1302 if (!ready) 1303 ready = kn->kn_data > 0; 1304 1305 return (ready); 1306 } 1307 1308 /*ARGSUSED*/ 1309 static int 1310 filt_pipewrite(struct knote *kn, long hint) 1311 { 1312 struct pipe *rpipe = (struct pipe *)kn->kn_fp->f_data; 1313 struct pipe *wpipe = rpipe->pipe_peer; 1314 int ready = 0; 1315 1316 kn->kn_data = 0; 1317 if (wpipe == NULL) { 1318 kn->kn_flags |= (EV_EOF | EV_NODATA); 1319 return (1); 1320 } 1321 1322 lwkt_gettoken(&wpipe->pipe_rlock); 1323 lwkt_gettoken(&wpipe->pipe_wlock); 1324 1325 if (wpipe->pipe_state & PIPE_WEOF) { 1326 kn->kn_flags |= (EV_EOF | EV_NODATA); 1327 ready = 1; 1328 } 1329 1330 if (!ready) 1331 kn->kn_data = wpipe->pipe_buffer.size - 1332 (wpipe->pipe_buffer.windex - 1333 wpipe->pipe_buffer.rindex); 1334 1335 lwkt_reltoken(&wpipe->pipe_wlock); 1336 lwkt_reltoken(&wpipe->pipe_rlock); 1337 1338 if (!ready) 1339 ready = kn->kn_data >= PIPE_BUF; 1340 1341 return (ready); 1342 } 1343