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