vfs_bio.c revision 1.269
1/* $NetBSD: vfs_bio.c,v 1.269 2017/01/20 09:45:13 skrll Exp $ */ 2 3/*- 4 * Copyright (c) 2007, 2008, 2009 The NetBSD Foundation, Inc. 5 * All rights reserved. 6 * 7 * This code is derived from software contributed to The NetBSD Foundation 8 * by Andrew Doran, and by Wasabi Systems, Inc. 9 * 10 * Redistribution and use in source and binary forms, with or without 11 * modification, are permitted provided that the following conditions 12 * are met: 13 * 1. Redistributions of source code must retain the above copyright 14 * notice, this list of conditions and the following disclaimer. 15 * 2. Redistributions in binary form must reproduce the above copyright 16 * notice, this list of conditions and the following disclaimer in the 17 * documentation and/or other materials provided with the distribution. 18 * 19 * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS 20 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED 21 * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR 22 * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS 23 * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR 24 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF 25 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS 26 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN 27 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) 28 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE 29 * POSSIBILITY OF SUCH DAMAGE. 30 */ 31 32/*- 33 * Copyright (c) 1982, 1986, 1989, 1993 34 * The Regents of the University of California. All rights reserved. 35 * (c) UNIX System Laboratories, Inc. 36 * All or some portions of this file are derived from material licensed 37 * to the University of California by American Telephone and Telegraph 38 * Co. or Unix System Laboratories, Inc. and are reproduced herein with 39 * the permission of UNIX System Laboratories, Inc. 40 * 41 * Redistribution and use in source and binary forms, with or without 42 * modification, are permitted provided that the following conditions 43 * are met: 44 * 1. Redistributions of source code must retain the above copyright 45 * notice, this list of conditions and the following disclaimer. 46 * 2. Redistributions in binary form must reproduce the above copyright 47 * notice, this list of conditions and the following disclaimer in the 48 * documentation and/or other materials provided with the distribution. 49 * 3. Neither the name of the University nor the names of its contributors 50 * may be used to endorse or promote products derived from this software 51 * without specific prior written permission. 52 * 53 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 54 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 55 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 56 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 57 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 58 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 59 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 60 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 61 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 62 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 63 * SUCH DAMAGE. 64 * 65 * @(#)vfs_bio.c 8.6 (Berkeley) 1/11/94 66 */ 67 68/*- 69 * Copyright (c) 1994 Christopher G. Demetriou 70 * 71 * Redistribution and use in source and binary forms, with or without 72 * modification, are permitted provided that the following conditions 73 * are met: 74 * 1. Redistributions of source code must retain the above copyright 75 * notice, this list of conditions and the following disclaimer. 76 * 2. Redistributions in binary form must reproduce the above copyright 77 * notice, this list of conditions and the following disclaimer in the 78 * documentation and/or other materials provided with the distribution. 79 * 3. All advertising materials mentioning features or use of this software 80 * must display the following acknowledgement: 81 * This product includes software developed by the University of 82 * California, Berkeley and its contributors. 83 * 4. Neither the name of the University nor the names of its contributors 84 * may be used to endorse or promote products derived from this software 85 * without specific prior written permission. 86 * 87 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 88 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 89 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 90 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 91 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 92 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 93 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 94 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 95 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 96 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 97 * SUCH DAMAGE. 98 * 99 * @(#)vfs_bio.c 8.6 (Berkeley) 1/11/94 100 */ 101 102/* 103 * The buffer cache subsystem. 104 * 105 * Some references: 106 * Bach: The Design of the UNIX Operating System (Prentice Hall, 1986) 107 * Leffler, et al.: The Design and Implementation of the 4.3BSD 108 * UNIX Operating System (Addison Welley, 1989) 109 * 110 * Locking 111 * 112 * There are three locks: 113 * - bufcache_lock: protects global buffer cache state. 114 * - BC_BUSY: a long term per-buffer lock. 115 * - buf_t::b_objlock: lock on completion (biowait vs biodone). 116 * 117 * For buffers associated with vnodes (a most common case) b_objlock points 118 * to the vnode_t::v_interlock. Otherwise, it points to generic buffer_lock. 119 * 120 * Lock order: 121 * bufcache_lock -> 122 * buf_t::b_objlock 123 */ 124 125#include <sys/cdefs.h> 126__KERNEL_RCSID(0, "$NetBSD: vfs_bio.c,v 1.269 2017/01/20 09:45:13 skrll Exp $"); 127 128#ifdef _KERNEL_OPT 129#include "opt_bufcache.h" 130#include "opt_dtrace.h" 131#include "opt_biohist.h" 132#endif 133 134#include <sys/param.h> 135#include <sys/systm.h> 136#include <sys/kernel.h> 137#include <sys/proc.h> 138#include <sys/buf.h> 139#include <sys/vnode.h> 140#include <sys/mount.h> 141#include <sys/resourcevar.h> 142#include <sys/sysctl.h> 143#include <sys/conf.h> 144#include <sys/kauth.h> 145#include <sys/fstrans.h> 146#include <sys/intr.h> 147#include <sys/cpu.h> 148#include <sys/wapbl.h> 149#include <sys/bitops.h> 150#include <sys/cprng.h> 151#include <sys/sdt.h> 152 153#include <uvm/uvm.h> /* extern struct uvm uvm */ 154 155#include <miscfs/specfs/specdev.h> 156 157#ifndef BUFPAGES 158# define BUFPAGES 0 159#endif 160 161#ifdef BUFCACHE 162# if (BUFCACHE < 5) || (BUFCACHE > 95) 163# error BUFCACHE is not between 5 and 95 164# endif 165#else 166# define BUFCACHE 15 167#endif 168 169u_int nbuf; /* desired number of buffer headers */ 170u_int bufpages = BUFPAGES; /* optional hardwired count */ 171u_int bufcache = BUFCACHE; /* max % of RAM to use for buffer cache */ 172 173/* Function prototypes */ 174struct bqueue; 175 176static void buf_setwm(void); 177static int buf_trim(void); 178static void *bufpool_page_alloc(struct pool *, int); 179static void bufpool_page_free(struct pool *, void *); 180static buf_t *bio_doread(struct vnode *, daddr_t, int, int); 181static buf_t *getnewbuf(int, int, int); 182static int buf_lotsfree(void); 183static int buf_canrelease(void); 184static u_long buf_mempoolidx(u_long); 185static u_long buf_roundsize(u_long); 186static void *buf_alloc(size_t); 187static void buf_mrelease(void *, size_t); 188static void binsheadfree(buf_t *, struct bqueue *); 189static void binstailfree(buf_t *, struct bqueue *); 190#ifdef DEBUG 191static int checkfreelist(buf_t *, struct bqueue *, int); 192#endif 193static void biointr(void *); 194static void biodone2(buf_t *); 195static void bref(buf_t *); 196static void brele(buf_t *); 197static void sysctl_kern_buf_setup(void); 198static void sysctl_vm_buf_setup(void); 199 200/* Initialization for biohist */ 201 202#include <sys/biohist.h> 203 204BIOHIST_DEFINE(biohist); 205 206void 207biohist_init(void) 208{ 209 210 BIOHIST_INIT(biohist, BIOHIST_SIZE); 211} 212 213/* 214 * Definitions for the buffer hash lists. 215 */ 216#define BUFHASH(dvp, lbn) \ 217 (&bufhashtbl[(((long)(dvp) >> 8) + (int)(lbn)) & bufhash]) 218LIST_HEAD(bufhashhdr, buf) *bufhashtbl, invalhash; 219u_long bufhash; 220struct bqueue bufqueues[BQUEUES]; 221 222static kcondvar_t needbuffer_cv; 223 224/* 225 * Buffer queue lock. 226 */ 227kmutex_t bufcache_lock; 228kmutex_t buffer_lock; 229 230/* Software ISR for completed transfers. */ 231static void *biodone_sih; 232 233/* Buffer pool for I/O buffers. */ 234static pool_cache_t buf_cache; 235static pool_cache_t bufio_cache; 236 237#define MEMPOOL_INDEX_OFFSET (ilog2(DEV_BSIZE)) /* smallest pool is 512 bytes */ 238#define NMEMPOOLS (ilog2(MAXBSIZE) - MEMPOOL_INDEX_OFFSET + 1) 239__CTASSERT((1 << (NMEMPOOLS + MEMPOOL_INDEX_OFFSET - 1)) == MAXBSIZE); 240 241/* Buffer memory pools */ 242static struct pool bmempools[NMEMPOOLS]; 243 244static struct vm_map *buf_map; 245 246/* 247 * Buffer memory pool allocator. 248 */ 249static void * 250bufpool_page_alloc(struct pool *pp, int flags) 251{ 252 253 return (void *)uvm_km_alloc(buf_map, 254 MAXBSIZE, MAXBSIZE, 255 ((flags & PR_WAITOK) ? 0 : UVM_KMF_NOWAIT|UVM_KMF_TRYLOCK) 256 | UVM_KMF_WIRED); 257} 258 259static void 260bufpool_page_free(struct pool *pp, void *v) 261{ 262 263 uvm_km_free(buf_map, (vaddr_t)v, MAXBSIZE, UVM_KMF_WIRED); 264} 265 266static struct pool_allocator bufmempool_allocator = { 267 .pa_alloc = bufpool_page_alloc, 268 .pa_free = bufpool_page_free, 269 .pa_pagesz = MAXBSIZE, 270}; 271 272/* Buffer memory management variables */ 273u_long bufmem_valimit; 274u_long bufmem_hiwater; 275u_long bufmem_lowater; 276u_long bufmem; 277 278/* 279 * MD code can call this to set a hard limit on the amount 280 * of virtual memory used by the buffer cache. 281 */ 282int 283buf_setvalimit(vsize_t sz) 284{ 285 286 /* We need to accommodate at least NMEMPOOLS of MAXBSIZE each */ 287 if (sz < NMEMPOOLS * MAXBSIZE) 288 return EINVAL; 289 290 bufmem_valimit = sz; 291 return 0; 292} 293 294static void 295buf_setwm(void) 296{ 297 298 bufmem_hiwater = buf_memcalc(); 299 /* lowater is approx. 2% of memory (with bufcache = 15) */ 300#define BUFMEM_WMSHIFT 3 301#define BUFMEM_HIWMMIN (64 * 1024 << BUFMEM_WMSHIFT) 302 if (bufmem_hiwater < BUFMEM_HIWMMIN) 303 /* Ensure a reasonable minimum value */ 304 bufmem_hiwater = BUFMEM_HIWMMIN; 305 bufmem_lowater = bufmem_hiwater >> BUFMEM_WMSHIFT; 306} 307 308#ifdef DEBUG 309int debug_verify_freelist = 0; 310static int 311checkfreelist(buf_t *bp, struct bqueue *dp, int ison) 312{ 313 buf_t *b; 314 315 if (!debug_verify_freelist) 316 return 1; 317 318 TAILQ_FOREACH(b, &dp->bq_queue, b_freelist) { 319 if (b == bp) 320 return ison ? 1 : 0; 321 } 322 323 return ison ? 0 : 1; 324} 325#endif 326 327/* 328 * Insq/Remq for the buffer hash lists. 329 * Call with buffer queue locked. 330 */ 331static void 332binsheadfree(buf_t *bp, struct bqueue *dp) 333{ 334 335 KASSERT(mutex_owned(&bufcache_lock)); 336 KASSERT(bp->b_freelistindex == -1); 337 TAILQ_INSERT_HEAD(&dp->bq_queue, bp, b_freelist); 338 dp->bq_bytes += bp->b_bufsize; 339 bp->b_freelistindex = dp - bufqueues; 340} 341 342static void 343binstailfree(buf_t *bp, struct bqueue *dp) 344{ 345 346 KASSERT(mutex_owned(&bufcache_lock)); 347 KASSERTMSG(bp->b_freelistindex == -1, "double free of buffer? " 348 "bp=%p, b_freelistindex=%d\n", bp, bp->b_freelistindex); 349 TAILQ_INSERT_TAIL(&dp->bq_queue, bp, b_freelist); 350 dp->bq_bytes += bp->b_bufsize; 351 bp->b_freelistindex = dp - bufqueues; 352} 353 354void 355bremfree(buf_t *bp) 356{ 357 struct bqueue *dp; 358 int bqidx = bp->b_freelistindex; 359 360 KASSERT(mutex_owned(&bufcache_lock)); 361 362 KASSERT(bqidx != -1); 363 dp = &bufqueues[bqidx]; 364 KDASSERT(checkfreelist(bp, dp, 1)); 365 KASSERT(dp->bq_bytes >= bp->b_bufsize); 366 TAILQ_REMOVE(&dp->bq_queue, bp, b_freelist); 367 dp->bq_bytes -= bp->b_bufsize; 368 369 /* For the sysctl helper. */ 370 if (bp == dp->bq_marker) 371 dp->bq_marker = NULL; 372 373#if defined(DIAGNOSTIC) 374 bp->b_freelistindex = -1; 375#endif /* defined(DIAGNOSTIC) */ 376} 377 378/* 379 * Add a reference to an buffer structure that came from buf_cache. 380 */ 381static inline void 382bref(buf_t *bp) 383{ 384 385 KASSERT(mutex_owned(&bufcache_lock)); 386 KASSERT(bp->b_refcnt > 0); 387 388 bp->b_refcnt++; 389} 390 391/* 392 * Free an unused buffer structure that came from buf_cache. 393 */ 394static inline void 395brele(buf_t *bp) 396{ 397 398 KASSERT(mutex_owned(&bufcache_lock)); 399 KASSERT(bp->b_refcnt > 0); 400 401 if (bp->b_refcnt-- == 1) { 402 buf_destroy(bp); 403#ifdef DEBUG 404 memset((char *)bp, 0, sizeof(*bp)); 405#endif 406 pool_cache_put(buf_cache, bp); 407 } 408} 409 410/* 411 * note that for some ports this is used by pmap bootstrap code to 412 * determine kva size. 413 */ 414u_long 415buf_memcalc(void) 416{ 417 u_long n; 418 vsize_t mapsz = 0; 419 420 /* 421 * Determine the upper bound of memory to use for buffers. 422 * 423 * - If bufpages is specified, use that as the number 424 * pages. 425 * 426 * - Otherwise, use bufcache as the percentage of 427 * physical memory. 428 */ 429 if (bufpages != 0) { 430 n = bufpages; 431 } else { 432 if (bufcache < 5) { 433 printf("forcing bufcache %d -> 5", bufcache); 434 bufcache = 5; 435 } 436 if (bufcache > 95) { 437 printf("forcing bufcache %d -> 95", bufcache); 438 bufcache = 95; 439 } 440 if (buf_map != NULL) 441 mapsz = vm_map_max(buf_map) - vm_map_min(buf_map); 442 n = calc_cache_size(mapsz, bufcache, 443 (buf_map != kernel_map) ? 100 : BUFCACHE_VA_MAXPCT) 444 / PAGE_SIZE; 445 } 446 447 n <<= PAGE_SHIFT; 448 if (bufmem_valimit != 0 && n > bufmem_valimit) 449 n = bufmem_valimit; 450 451 return (n); 452} 453 454/* 455 * Initialize buffers and hash links for buffers. 456 */ 457void 458bufinit(void) 459{ 460 struct bqueue *dp; 461 int use_std; 462 u_int i; 463 464 biodone_vfs = biodone; 465 466 mutex_init(&bufcache_lock, MUTEX_DEFAULT, IPL_NONE); 467 mutex_init(&buffer_lock, MUTEX_DEFAULT, IPL_NONE); 468 cv_init(&needbuffer_cv, "needbuf"); 469 470 if (bufmem_valimit != 0) { 471 vaddr_t minaddr = 0, maxaddr; 472 buf_map = uvm_km_suballoc(kernel_map, &minaddr, &maxaddr, 473 bufmem_valimit, 0, false, 0); 474 if (buf_map == NULL) 475 panic("bufinit: cannot allocate submap"); 476 } else 477 buf_map = kernel_map; 478 479 /* 480 * Initialize buffer cache memory parameters. 481 */ 482 bufmem = 0; 483 buf_setwm(); 484 485 /* On "small" machines use small pool page sizes where possible */ 486 use_std = (physmem < atop(16*1024*1024)); 487 488 /* 489 * Also use them on systems that can map the pool pages using 490 * a direct-mapped segment. 491 */ 492#ifdef PMAP_MAP_POOLPAGE 493 use_std = 1; 494#endif 495 496 buf_cache = pool_cache_init(sizeof(buf_t), 0, 0, 0, 497 "bufpl", NULL, IPL_SOFTBIO, NULL, NULL, NULL); 498 bufio_cache = pool_cache_init(sizeof(buf_t), 0, 0, 0, 499 "biopl", NULL, IPL_BIO, NULL, NULL, NULL); 500 501 for (i = 0; i < NMEMPOOLS; i++) { 502 struct pool_allocator *pa; 503 struct pool *pp = &bmempools[i]; 504 u_int size = 1 << (i + MEMPOOL_INDEX_OFFSET); 505 char *name = kmem_alloc(8, KM_SLEEP); /* XXX: never freed */ 506 if (__predict_false(size >= 1048576)) 507 (void)snprintf(name, 8, "buf%um", size / 1048576); 508 else if (__predict_true(size >= 1024)) 509 (void)snprintf(name, 8, "buf%uk", size / 1024); 510 else 511 (void)snprintf(name, 8, "buf%ub", size); 512 pa = (size <= PAGE_SIZE && use_std) 513 ? &pool_allocator_nointr 514 : &bufmempool_allocator; 515 pool_init(pp, size, 0, 0, 0, name, pa, IPL_NONE); 516 pool_setlowat(pp, 1); 517 pool_sethiwat(pp, 1); 518 } 519 520 /* Initialize the buffer queues */ 521 for (dp = bufqueues; dp < &bufqueues[BQUEUES]; dp++) { 522 TAILQ_INIT(&dp->bq_queue); 523 dp->bq_bytes = 0; 524 } 525 526 /* 527 * Estimate hash table size based on the amount of memory we 528 * intend to use for the buffer cache. The average buffer 529 * size is dependent on our clients (i.e. filesystems). 530 * 531 * For now, use an empirical 3K per buffer. 532 */ 533 nbuf = (bufmem_hiwater / 1024) / 3; 534 bufhashtbl = hashinit(nbuf, HASH_LIST, true, &bufhash); 535 536 sysctl_kern_buf_setup(); 537 sysctl_vm_buf_setup(); 538} 539 540void 541bufinit2(void) 542{ 543 544 biodone_sih = softint_establish(SOFTINT_BIO | SOFTINT_MPSAFE, biointr, 545 NULL); 546 if (biodone_sih == NULL) 547 panic("bufinit2: can't establish soft interrupt"); 548} 549 550static int 551buf_lotsfree(void) 552{ 553 u_long guess; 554 555 /* Always allocate if less than the low water mark. */ 556 if (bufmem < bufmem_lowater) 557 return 1; 558 559 /* Never allocate if greater than the high water mark. */ 560 if (bufmem > bufmem_hiwater) 561 return 0; 562 563 /* If there's anything on the AGE list, it should be eaten. */ 564 if (TAILQ_FIRST(&bufqueues[BQ_AGE].bq_queue) != NULL) 565 return 0; 566 567 /* 568 * The probabily of getting a new allocation is inversely 569 * proportional to the current size of the cache above 570 * the low water mark. Divide the total first to avoid overflows 571 * in the product. 572 */ 573 guess = cprng_fast32() % 16; 574 575 if ((bufmem_hiwater - bufmem_lowater) / 16 * guess >= 576 (bufmem - bufmem_lowater)) 577 return 1; 578 579 /* Otherwise don't allocate. */ 580 return 0; 581} 582 583/* 584 * Return estimate of bytes we think need to be 585 * released to help resolve low memory conditions. 586 * 587 * => called with bufcache_lock held. 588 */ 589static int 590buf_canrelease(void) 591{ 592 int pagedemand, ninvalid = 0; 593 594 KASSERT(mutex_owned(&bufcache_lock)); 595 596 if (bufmem < bufmem_lowater) 597 return 0; 598 599 if (bufmem > bufmem_hiwater) 600 return bufmem - bufmem_hiwater; 601 602 ninvalid += bufqueues[BQ_AGE].bq_bytes; 603 604 pagedemand = uvmexp.freetarg - uvmexp.free; 605 if (pagedemand < 0) 606 return ninvalid; 607 return MAX(ninvalid, MIN(2 * MAXBSIZE, 608 MIN((bufmem - bufmem_lowater) / 16, pagedemand * PAGE_SIZE))); 609} 610 611/* 612 * Buffer memory allocation helper functions 613 */ 614static u_long 615buf_mempoolidx(u_long size) 616{ 617 u_int n = 0; 618 619 size -= 1; 620 size >>= MEMPOOL_INDEX_OFFSET; 621 while (size) { 622 size >>= 1; 623 n += 1; 624 } 625 if (n >= NMEMPOOLS) 626 panic("buf mem pool index %d", n); 627 return n; 628} 629 630static u_long 631buf_roundsize(u_long size) 632{ 633 /* Round up to nearest power of 2 */ 634 return (1 << (buf_mempoolidx(size) + MEMPOOL_INDEX_OFFSET)); 635} 636 637static void * 638buf_alloc(size_t size) 639{ 640 u_int n = buf_mempoolidx(size); 641 void *addr; 642 643 while (1) { 644 addr = pool_get(&bmempools[n], PR_NOWAIT); 645 if (addr != NULL) 646 break; 647 648 /* No memory, see if we can free some. If so, try again */ 649 mutex_enter(&bufcache_lock); 650 if (buf_drain(1) > 0) { 651 mutex_exit(&bufcache_lock); 652 continue; 653 } 654 655 if (curlwp == uvm.pagedaemon_lwp) { 656 mutex_exit(&bufcache_lock); 657 return NULL; 658 } 659 660 /* Wait for buffers to arrive on the LRU queue */ 661 cv_timedwait(&needbuffer_cv, &bufcache_lock, hz / 4); 662 mutex_exit(&bufcache_lock); 663 } 664 665 return addr; 666} 667 668static void 669buf_mrelease(void *addr, size_t size) 670{ 671 672 pool_put(&bmempools[buf_mempoolidx(size)], addr); 673} 674 675/* 676 * bread()/breadn() helper. 677 */ 678static buf_t * 679bio_doread(struct vnode *vp, daddr_t blkno, int size, int async) 680{ 681 buf_t *bp; 682 struct mount *mp; 683 684 bp = getblk(vp, blkno, size, 0, 0); 685 686 /* 687 * getblk() may return NULL if we are the pagedaemon. 688 */ 689 if (bp == NULL) { 690 KASSERT(curlwp == uvm.pagedaemon_lwp); 691 return NULL; 692 } 693 694 /* 695 * If buffer does not have data valid, start a read. 696 * Note that if buffer is BC_INVAL, getblk() won't return it. 697 * Therefore, it's valid if its I/O has completed or been delayed. 698 */ 699 if (!ISSET(bp->b_oflags, (BO_DONE | BO_DELWRI))) { 700 /* Start I/O for the buffer. */ 701 SET(bp->b_flags, B_READ | async); 702 if (async) 703 BIO_SETPRIO(bp, BPRIO_TIMELIMITED); 704 else 705 BIO_SETPRIO(bp, BPRIO_TIMECRITICAL); 706 VOP_STRATEGY(vp, bp); 707 708 /* Pay for the read. */ 709 curlwp->l_ru.ru_inblock++; 710 } else if (async) 711 brelse(bp, 0); 712 713 if (vp->v_type == VBLK) 714 mp = spec_node_getmountedfs(vp); 715 else 716 mp = vp->v_mount; 717 718 /* 719 * Collect statistics on synchronous and asynchronous reads. 720 * Reads from block devices are charged to their associated 721 * filesystem (if any). 722 */ 723 if (mp != NULL) { 724 if (async == 0) 725 mp->mnt_stat.f_syncreads++; 726 else 727 mp->mnt_stat.f_asyncreads++; 728 } 729 730 return (bp); 731} 732 733/* 734 * Read a disk block. 735 * This algorithm described in Bach (p.54). 736 */ 737int 738bread(struct vnode *vp, daddr_t blkno, int size, int flags, buf_t **bpp) 739{ 740 buf_t *bp; 741 int error; 742 743 BIOHIST_FUNC(__func__); BIOHIST_CALLED(biohist); 744 745 /* Get buffer for block. */ 746 bp = *bpp = bio_doread(vp, blkno, size, 0); 747 if (bp == NULL) 748 return ENOMEM; 749 750 /* Wait for the read to complete, and return result. */ 751 error = biowait(bp); 752 if (error == 0 && (flags & B_MODIFY) != 0) 753 error = fscow_run(bp, true); 754 if (error) { 755 brelse(bp, 0); 756 *bpp = NULL; 757 } 758 759 return error; 760} 761 762/* 763 * Read-ahead multiple disk blocks. The first is sync, the rest async. 764 * Trivial modification to the breada algorithm presented in Bach (p.55). 765 */ 766int 767breadn(struct vnode *vp, daddr_t blkno, int size, daddr_t *rablks, 768 int *rasizes, int nrablks, int flags, buf_t **bpp) 769{ 770 buf_t *bp; 771 int error, i; 772 773 BIOHIST_FUNC(__func__); BIOHIST_CALLED(biohist); 774 775 bp = *bpp = bio_doread(vp, blkno, size, 0); 776 if (bp == NULL) 777 return ENOMEM; 778 779 /* 780 * For each of the read-ahead blocks, start a read, if necessary. 781 */ 782 mutex_enter(&bufcache_lock); 783 for (i = 0; i < nrablks; i++) { 784 /* If it's in the cache, just go on to next one. */ 785 if (incore(vp, rablks[i])) 786 continue; 787 788 /* Get a buffer for the read-ahead block */ 789 mutex_exit(&bufcache_lock); 790 (void) bio_doread(vp, rablks[i], rasizes[i], B_ASYNC); 791 mutex_enter(&bufcache_lock); 792 } 793 mutex_exit(&bufcache_lock); 794 795 /* Otherwise, we had to start a read for it; wait until it's valid. */ 796 error = biowait(bp); 797 if (error == 0 && (flags & B_MODIFY) != 0) 798 error = fscow_run(bp, true); 799 if (error) { 800 brelse(bp, 0); 801 *bpp = NULL; 802 } 803 804 return error; 805} 806 807/* 808 * Block write. Described in Bach (p.56) 809 */ 810int 811bwrite(buf_t *bp) 812{ 813 int rv, sync, wasdelayed; 814 struct vnode *vp; 815 struct mount *mp; 816 817 BIOHIST_FUNC(__func__); BIOHIST_CALLARGS(biohist, "bp=%p", 818 bp, 0, 0, 0); 819 820 KASSERT(ISSET(bp->b_cflags, BC_BUSY)); 821 KASSERT(!cv_has_waiters(&bp->b_done)); 822 823 vp = bp->b_vp; 824 825 /* 826 * dholland 20160728 AFAICT vp==NULL must be impossible as it 827 * will crash upon reaching VOP_STRATEGY below... see further 828 * analysis on tech-kern. 829 */ 830 KASSERTMSG(vp != NULL, "bwrite given buffer with null vnode"); 831 832 if (vp != NULL) { 833 KASSERT(bp->b_objlock == vp->v_interlock); 834 if (vp->v_type == VBLK) 835 mp = spec_node_getmountedfs(vp); 836 else 837 mp = vp->v_mount; 838 } else { 839 mp = NULL; 840 } 841 842 if (mp && mp->mnt_wapbl) { 843 if (bp->b_iodone != mp->mnt_wapbl_op->wo_wapbl_biodone) { 844 bdwrite(bp); 845 return 0; 846 } 847 } 848 849 /* 850 * Remember buffer type, to switch on it later. If the write was 851 * synchronous, but the file system was mounted with MNT_ASYNC, 852 * convert it to a delayed write. 853 * XXX note that this relies on delayed tape writes being converted 854 * to async, not sync writes (which is safe, but ugly). 855 */ 856 sync = !ISSET(bp->b_flags, B_ASYNC); 857 if (sync && mp != NULL && ISSET(mp->mnt_flag, MNT_ASYNC)) { 858 bdwrite(bp); 859 return (0); 860 } 861 862 /* 863 * Collect statistics on synchronous and asynchronous writes. 864 * Writes to block devices are charged to their associated 865 * filesystem (if any). 866 */ 867 if (mp != NULL) { 868 if (sync) 869 mp->mnt_stat.f_syncwrites++; 870 else 871 mp->mnt_stat.f_asyncwrites++; 872 } 873 874 /* 875 * Pay for the I/O operation and make sure the buf is on the correct 876 * vnode queue. 877 */ 878 bp->b_error = 0; 879 wasdelayed = ISSET(bp->b_oflags, BO_DELWRI); 880 CLR(bp->b_flags, B_READ); 881 if (wasdelayed) { 882 mutex_enter(&bufcache_lock); 883 mutex_enter(bp->b_objlock); 884 CLR(bp->b_oflags, BO_DONE | BO_DELWRI); 885 reassignbuf(bp, bp->b_vp); 886 mutex_exit(&bufcache_lock); 887 } else { 888 curlwp->l_ru.ru_oublock++; 889 mutex_enter(bp->b_objlock); 890 CLR(bp->b_oflags, BO_DONE | BO_DELWRI); 891 } 892 if (vp != NULL) 893 vp->v_numoutput++; 894 mutex_exit(bp->b_objlock); 895 896 /* Initiate disk write. */ 897 if (sync) 898 BIO_SETPRIO(bp, BPRIO_TIMECRITICAL); 899 else 900 BIO_SETPRIO(bp, BPRIO_TIMELIMITED); 901 902 VOP_STRATEGY(vp, bp); 903 904 if (sync) { 905 /* If I/O was synchronous, wait for it to complete. */ 906 rv = biowait(bp); 907 908 /* Release the buffer. */ 909 brelse(bp, 0); 910 911 return (rv); 912 } else { 913 return (0); 914 } 915} 916 917int 918vn_bwrite(void *v) 919{ 920 struct vop_bwrite_args *ap = v; 921 922 return (bwrite(ap->a_bp)); 923} 924 925/* 926 * Delayed write. 927 * 928 * The buffer is marked dirty, but is not queued for I/O. 929 * This routine should be used when the buffer is expected 930 * to be modified again soon, typically a small write that 931 * partially fills a buffer. 932 * 933 * NB: magnetic tapes cannot be delayed; they must be 934 * written in the order that the writes are requested. 935 * 936 * Described in Leffler, et al. (pp. 208-213). 937 */ 938void 939bdwrite(buf_t *bp) 940{ 941 942 BIOHIST_FUNC(__func__); BIOHIST_CALLARGS(biohist, "bp=%p", 943 bp, 0, 0, 0); 944 945 KASSERT(bp->b_vp == NULL || bp->b_vp->v_tag != VT_UFS || 946 bp->b_vp->v_type == VBLK || ISSET(bp->b_flags, B_COWDONE)); 947 KASSERT(ISSET(bp->b_cflags, BC_BUSY)); 948 KASSERT(!cv_has_waiters(&bp->b_done)); 949 950 /* If this is a tape block, write the block now. */ 951 if (bdev_type(bp->b_dev) == D_TAPE) { 952 bawrite(bp); 953 return; 954 } 955 956 if (wapbl_vphaswapbl(bp->b_vp)) { 957 struct mount *mp = wapbl_vptomp(bp->b_vp); 958 959 if (bp->b_iodone != mp->mnt_wapbl_op->wo_wapbl_biodone) { 960 WAPBL_ADD_BUF(mp, bp); 961 } 962 } 963 964 /* 965 * If the block hasn't been seen before: 966 * (1) Mark it as having been seen, 967 * (2) Charge for the write, 968 * (3) Make sure it's on its vnode's correct block list. 969 */ 970 KASSERT(bp->b_vp == NULL || bp->b_objlock == bp->b_vp->v_interlock); 971 972 if (!ISSET(bp->b_oflags, BO_DELWRI)) { 973 mutex_enter(&bufcache_lock); 974 mutex_enter(bp->b_objlock); 975 SET(bp->b_oflags, BO_DELWRI); 976 curlwp->l_ru.ru_oublock++; 977 reassignbuf(bp, bp->b_vp); 978 mutex_exit(&bufcache_lock); 979 } else { 980 mutex_enter(bp->b_objlock); 981 } 982 /* Otherwise, the "write" is done, so mark and release the buffer. */ 983 CLR(bp->b_oflags, BO_DONE); 984 mutex_exit(bp->b_objlock); 985 986 brelse(bp, 0); 987} 988 989/* 990 * Asynchronous block write; just an asynchronous bwrite(). 991 */ 992void 993bawrite(buf_t *bp) 994{ 995 996 KASSERT(ISSET(bp->b_cflags, BC_BUSY)); 997 KASSERT(bp->b_vp != NULL); 998 999 SET(bp->b_flags, B_ASYNC); 1000 VOP_BWRITE(bp->b_vp, bp); 1001} 1002 1003/* 1004 * Release a buffer on to the free lists. 1005 * Described in Bach (p. 46). 1006 */ 1007void 1008brelsel(buf_t *bp, int set) 1009{ 1010 struct bqueue *bufq; 1011 struct vnode *vp; 1012 1013 KASSERT(bp != NULL); 1014 KASSERT(mutex_owned(&bufcache_lock)); 1015 KASSERT(!cv_has_waiters(&bp->b_done)); 1016 KASSERT(bp->b_refcnt > 0); 1017 1018 SET(bp->b_cflags, set); 1019 1020 KASSERT(ISSET(bp->b_cflags, BC_BUSY)); 1021 KASSERT(bp->b_iodone == NULL); 1022 1023 /* Wake up any processes waiting for any buffer to become free. */ 1024 cv_signal(&needbuffer_cv); 1025 1026 /* Wake up any proceeses waiting for _this_ buffer to become free */ 1027 if (ISSET(bp->b_cflags, BC_WANTED)) 1028 CLR(bp->b_cflags, BC_WANTED|BC_AGE); 1029 1030 /* If it's clean clear the copy-on-write flag. */ 1031 if (ISSET(bp->b_flags, B_COWDONE)) { 1032 mutex_enter(bp->b_objlock); 1033 if (!ISSET(bp->b_oflags, BO_DELWRI)) 1034 CLR(bp->b_flags, B_COWDONE); 1035 mutex_exit(bp->b_objlock); 1036 } 1037 1038 /* 1039 * Determine which queue the buffer should be on, then put it there. 1040 */ 1041 1042 /* If it's locked, don't report an error; try again later. */ 1043 if (ISSET(bp->b_flags, B_LOCKED)) 1044 bp->b_error = 0; 1045 1046 /* If it's not cacheable, or an error, mark it invalid. */ 1047 if (ISSET(bp->b_cflags, BC_NOCACHE) || bp->b_error != 0) 1048 SET(bp->b_cflags, BC_INVAL); 1049 1050 if (ISSET(bp->b_cflags, BC_VFLUSH)) { 1051 /* 1052 * This is a delayed write buffer that was just flushed to 1053 * disk. It is still on the LRU queue. If it's become 1054 * invalid, then we need to move it to a different queue; 1055 * otherwise leave it in its current position. 1056 */ 1057 CLR(bp->b_cflags, BC_VFLUSH); 1058 if (!ISSET(bp->b_cflags, BC_INVAL|BC_AGE) && 1059 !ISSET(bp->b_flags, B_LOCKED) && bp->b_error == 0) { 1060 KDASSERT(checkfreelist(bp, &bufqueues[BQ_LRU], 1)); 1061 goto already_queued; 1062 } else { 1063 bremfree(bp); 1064 } 1065 } 1066 1067 KDASSERT(checkfreelist(bp, &bufqueues[BQ_AGE], 0)); 1068 KDASSERT(checkfreelist(bp, &bufqueues[BQ_LRU], 0)); 1069 KDASSERT(checkfreelist(bp, &bufqueues[BQ_LOCKED], 0)); 1070 1071 if ((bp->b_bufsize <= 0) || ISSET(bp->b_cflags, BC_INVAL)) { 1072 /* 1073 * If it's invalid or empty, dissociate it from its vnode 1074 * and put on the head of the appropriate queue. 1075 */ 1076 if (ISSET(bp->b_flags, B_LOCKED)) { 1077 if (wapbl_vphaswapbl(vp = bp->b_vp)) { 1078 struct mount *mp = wapbl_vptomp(vp); 1079 1080 KASSERT(bp->b_iodone 1081 != mp->mnt_wapbl_op->wo_wapbl_biodone); 1082 WAPBL_REMOVE_BUF(mp, bp); 1083 } 1084 } 1085 1086 mutex_enter(bp->b_objlock); 1087 CLR(bp->b_oflags, BO_DONE|BO_DELWRI); 1088 if ((vp = bp->b_vp) != NULL) { 1089 KASSERT(bp->b_objlock == vp->v_interlock); 1090 reassignbuf(bp, bp->b_vp); 1091 brelvp(bp); 1092 mutex_exit(vp->v_interlock); 1093 } else { 1094 KASSERT(bp->b_objlock == &buffer_lock); 1095 mutex_exit(bp->b_objlock); 1096 } 1097 1098 if (bp->b_bufsize <= 0) 1099 /* no data */ 1100 goto already_queued; 1101 else 1102 /* invalid data */ 1103 bufq = &bufqueues[BQ_AGE]; 1104 binsheadfree(bp, bufq); 1105 } else { 1106 /* 1107 * It has valid data. Put it on the end of the appropriate 1108 * queue, so that it'll stick around for as long as possible. 1109 * If buf is AGE, but has dependencies, must put it on last 1110 * bufqueue to be scanned, ie LRU. This protects against the 1111 * livelock where BQ_AGE only has buffers with dependencies, 1112 * and we thus never get to the dependent buffers in BQ_LRU. 1113 */ 1114 if (ISSET(bp->b_flags, B_LOCKED)) { 1115 /* locked in core */ 1116 bufq = &bufqueues[BQ_LOCKED]; 1117 } else if (!ISSET(bp->b_cflags, BC_AGE)) { 1118 /* valid data */ 1119 bufq = &bufqueues[BQ_LRU]; 1120 } else { 1121 /* stale but valid data */ 1122 bufq = &bufqueues[BQ_AGE]; 1123 } 1124 binstailfree(bp, bufq); 1125 } 1126already_queued: 1127 /* Unlock the buffer. */ 1128 CLR(bp->b_cflags, BC_AGE|BC_BUSY|BC_NOCACHE); 1129 CLR(bp->b_flags, B_ASYNC); 1130 cv_broadcast(&bp->b_busy); 1131 1132 if (bp->b_bufsize <= 0) 1133 brele(bp); 1134} 1135 1136void 1137brelse(buf_t *bp, int set) 1138{ 1139 1140 mutex_enter(&bufcache_lock); 1141 brelsel(bp, set); 1142 mutex_exit(&bufcache_lock); 1143} 1144 1145/* 1146 * Determine if a block is in the cache. 1147 * Just look on what would be its hash chain. If it's there, return 1148 * a pointer to it, unless it's marked invalid. If it's marked invalid, 1149 * we normally don't return the buffer, unless the caller explicitly 1150 * wants us to. 1151 */ 1152buf_t * 1153incore(struct vnode *vp, daddr_t blkno) 1154{ 1155 buf_t *bp; 1156 1157 KASSERT(mutex_owned(&bufcache_lock)); 1158 1159 /* Search hash chain */ 1160 LIST_FOREACH(bp, BUFHASH(vp, blkno), b_hash) { 1161 if (bp->b_lblkno == blkno && bp->b_vp == vp && 1162 !ISSET(bp->b_cflags, BC_INVAL)) { 1163 KASSERT(bp->b_objlock == vp->v_interlock); 1164 return (bp); 1165 } 1166 } 1167 1168 return (NULL); 1169} 1170 1171/* 1172 * Get a block of requested size that is associated with 1173 * a given vnode and block offset. If it is found in the 1174 * block cache, mark it as having been found, make it busy 1175 * and return it. Otherwise, return an empty block of the 1176 * correct size. It is up to the caller to insure that the 1177 * cached blocks be of the correct size. 1178 */ 1179buf_t * 1180getblk(struct vnode *vp, daddr_t blkno, int size, int slpflag, int slptimeo) 1181{ 1182 int err, preserve; 1183 buf_t *bp; 1184 1185 mutex_enter(&bufcache_lock); 1186 loop: 1187 bp = incore(vp, blkno); 1188 if (bp != NULL) { 1189 err = bbusy(bp, ((slpflag & PCATCH) != 0), slptimeo, NULL); 1190 if (err != 0) { 1191 if (err == EPASSTHROUGH) 1192 goto loop; 1193 mutex_exit(&bufcache_lock); 1194 return (NULL); 1195 } 1196 KASSERT(!cv_has_waiters(&bp->b_done)); 1197#ifdef DIAGNOSTIC 1198 if (ISSET(bp->b_oflags, BO_DONE|BO_DELWRI) && 1199 bp->b_bcount < size && vp->v_type != VBLK) 1200 panic("getblk: block size invariant failed"); 1201#endif 1202 bremfree(bp); 1203 preserve = 1; 1204 } else { 1205 if ((bp = getnewbuf(slpflag, slptimeo, 0)) == NULL) 1206 goto loop; 1207 1208 if (incore(vp, blkno) != NULL) { 1209 /* The block has come into memory in the meantime. */ 1210 brelsel(bp, 0); 1211 goto loop; 1212 } 1213 1214 LIST_INSERT_HEAD(BUFHASH(vp, blkno), bp, b_hash); 1215 bp->b_blkno = bp->b_lblkno = bp->b_rawblkno = blkno; 1216 mutex_enter(vp->v_interlock); 1217 bgetvp(vp, bp); 1218 mutex_exit(vp->v_interlock); 1219 preserve = 0; 1220 } 1221 mutex_exit(&bufcache_lock); 1222 1223 /* 1224 * LFS can't track total size of B_LOCKED buffer (locked_queue_bytes) 1225 * if we re-size buffers here. 1226 */ 1227 if (ISSET(bp->b_flags, B_LOCKED)) { 1228 KASSERT(bp->b_bufsize >= size); 1229 } else { 1230 if (allocbuf(bp, size, preserve)) { 1231 mutex_enter(&bufcache_lock); 1232 LIST_REMOVE(bp, b_hash); 1233 mutex_exit(&bufcache_lock); 1234 brelse(bp, BC_INVAL); 1235 return NULL; 1236 } 1237 } 1238 BIO_SETPRIO(bp, BPRIO_DEFAULT); 1239 return (bp); 1240} 1241 1242/* 1243 * Get an empty, disassociated buffer of given size. 1244 */ 1245buf_t * 1246geteblk(int size) 1247{ 1248 buf_t *bp; 1249 int error __diagused; 1250 1251 mutex_enter(&bufcache_lock); 1252 while ((bp = getnewbuf(0, 0, 0)) == NULL) 1253 ; 1254 1255 SET(bp->b_cflags, BC_INVAL); 1256 LIST_INSERT_HEAD(&invalhash, bp, b_hash); 1257 mutex_exit(&bufcache_lock); 1258 BIO_SETPRIO(bp, BPRIO_DEFAULT); 1259 error = allocbuf(bp, size, 0); 1260 KASSERT(error == 0); 1261 return (bp); 1262} 1263 1264/* 1265 * Expand or contract the actual memory allocated to a buffer. 1266 * 1267 * If the buffer shrinks, data is lost, so it's up to the 1268 * caller to have written it out *first*; this routine will not 1269 * start a write. If the buffer grows, it's the callers 1270 * responsibility to fill out the buffer's additional contents. 1271 */ 1272int 1273allocbuf(buf_t *bp, int size, int preserve) 1274{ 1275 void *addr; 1276 vsize_t oldsize, desired_size; 1277 int oldcount; 1278 int delta; 1279 1280 desired_size = buf_roundsize(size); 1281 if (desired_size > MAXBSIZE) 1282 printf("allocbuf: buffer larger than MAXBSIZE requested"); 1283 1284 oldcount = bp->b_bcount; 1285 1286 bp->b_bcount = size; 1287 1288 oldsize = bp->b_bufsize; 1289 if (oldsize == desired_size) { 1290 /* 1291 * Do not short cut the WAPBL resize, as the buffer length 1292 * could still have changed and this would corrupt the 1293 * tracking of the transaction length. 1294 */ 1295 goto out; 1296 } 1297 1298 /* 1299 * If we want a buffer of a different size, re-allocate the 1300 * buffer's memory; copy old content only if needed. 1301 */ 1302 addr = buf_alloc(desired_size); 1303 if (addr == NULL) 1304 return ENOMEM; 1305 if (preserve) 1306 memcpy(addr, bp->b_data, MIN(oldsize,desired_size)); 1307 if (bp->b_data != NULL) 1308 buf_mrelease(bp->b_data, oldsize); 1309 bp->b_data = addr; 1310 bp->b_bufsize = desired_size; 1311 1312 /* 1313 * Update overall buffer memory counter (protected by bufcache_lock) 1314 */ 1315 delta = (long)desired_size - (long)oldsize; 1316 1317 mutex_enter(&bufcache_lock); 1318 if ((bufmem += delta) > bufmem_hiwater) { 1319 /* 1320 * Need to trim overall memory usage. 1321 */ 1322 while (buf_canrelease()) { 1323 if (curcpu()->ci_schedstate.spc_flags & 1324 SPCF_SHOULDYIELD) { 1325 mutex_exit(&bufcache_lock); 1326 preempt(); 1327 mutex_enter(&bufcache_lock); 1328 } 1329 if (buf_trim() == 0) 1330 break; 1331 } 1332 } 1333 mutex_exit(&bufcache_lock); 1334 1335 out: 1336 if (wapbl_vphaswapbl(bp->b_vp)) 1337 WAPBL_RESIZE_BUF(wapbl_vptomp(bp->b_vp), bp, oldsize, oldcount); 1338 1339 return 0; 1340} 1341 1342/* 1343 * Find a buffer which is available for use. 1344 * Select something from a free list. 1345 * Preference is to AGE list, then LRU list. 1346 * 1347 * Called with the buffer queues locked. 1348 * Return buffer locked. 1349 */ 1350buf_t * 1351getnewbuf(int slpflag, int slptimeo, int from_bufq) 1352{ 1353 buf_t *bp; 1354 struct vnode *vp; 1355 1356 start: 1357 KASSERT(mutex_owned(&bufcache_lock)); 1358 1359 /* 1360 * Get a new buffer from the pool. 1361 */ 1362 if (!from_bufq && buf_lotsfree()) { 1363 mutex_exit(&bufcache_lock); 1364 bp = pool_cache_get(buf_cache, PR_NOWAIT); 1365 if (bp != NULL) { 1366 memset((char *)bp, 0, sizeof(*bp)); 1367 buf_init(bp); 1368 SET(bp->b_cflags, BC_BUSY); /* mark buffer busy */ 1369 mutex_enter(&bufcache_lock); 1370#if defined(DIAGNOSTIC) 1371 bp->b_freelistindex = -1; 1372#endif /* defined(DIAGNOSTIC) */ 1373 return (bp); 1374 } 1375 mutex_enter(&bufcache_lock); 1376 } 1377 1378 KASSERT(mutex_owned(&bufcache_lock)); 1379 if ((bp = TAILQ_FIRST(&bufqueues[BQ_AGE].bq_queue)) != NULL || 1380 (bp = TAILQ_FIRST(&bufqueues[BQ_LRU].bq_queue)) != NULL) { 1381 KASSERT(!ISSET(bp->b_cflags, BC_BUSY) || ISSET(bp->b_cflags, BC_VFLUSH)); 1382 bremfree(bp); 1383 1384 /* Buffer is no longer on free lists. */ 1385 SET(bp->b_cflags, BC_BUSY); 1386 } else { 1387 /* 1388 * XXX: !from_bufq should be removed. 1389 */ 1390 if (!from_bufq || curlwp != uvm.pagedaemon_lwp) { 1391 /* wait for a free buffer of any kind */ 1392 if ((slpflag & PCATCH) != 0) 1393 (void)cv_timedwait_sig(&needbuffer_cv, 1394 &bufcache_lock, slptimeo); 1395 else 1396 (void)cv_timedwait(&needbuffer_cv, 1397 &bufcache_lock, slptimeo); 1398 } 1399 return (NULL); 1400 } 1401 1402#ifdef DIAGNOSTIC 1403 if (bp->b_bufsize <= 0) 1404 panic("buffer %p: on queue but empty", bp); 1405#endif 1406 1407 if (ISSET(bp->b_cflags, BC_VFLUSH)) { 1408 /* 1409 * This is a delayed write buffer being flushed to disk. Make 1410 * sure it gets aged out of the queue when it's finished, and 1411 * leave it off the LRU queue. 1412 */ 1413 CLR(bp->b_cflags, BC_VFLUSH); 1414 SET(bp->b_cflags, BC_AGE); 1415 goto start; 1416 } 1417 1418 KASSERT(ISSET(bp->b_cflags, BC_BUSY)); 1419 KASSERT(bp->b_refcnt > 0); 1420 KASSERT(!cv_has_waiters(&bp->b_done)); 1421 1422 /* 1423 * If buffer was a delayed write, start it and return NULL 1424 * (since we might sleep while starting the write). 1425 */ 1426 if (ISSET(bp->b_oflags, BO_DELWRI)) { 1427 /* 1428 * This buffer has gone through the LRU, so make sure it gets 1429 * reused ASAP. 1430 */ 1431 SET(bp->b_cflags, BC_AGE); 1432 mutex_exit(&bufcache_lock); 1433 bawrite(bp); 1434 mutex_enter(&bufcache_lock); 1435 return (NULL); 1436 } 1437 1438 vp = bp->b_vp; 1439 1440 /* clear out various other fields */ 1441 bp->b_cflags = BC_BUSY; 1442 bp->b_oflags = 0; 1443 bp->b_flags = 0; 1444 bp->b_dev = NODEV; 1445 bp->b_blkno = 0; 1446 bp->b_lblkno = 0; 1447 bp->b_rawblkno = 0; 1448 bp->b_iodone = 0; 1449 bp->b_error = 0; 1450 bp->b_resid = 0; 1451 bp->b_bcount = 0; 1452 1453 LIST_REMOVE(bp, b_hash); 1454 1455 /* Disassociate us from our vnode, if we had one... */ 1456 if (vp != NULL) { 1457 mutex_enter(vp->v_interlock); 1458 brelvp(bp); 1459 mutex_exit(vp->v_interlock); 1460 } 1461 1462 return (bp); 1463} 1464 1465/* 1466 * Attempt to free an aged buffer off the queues. 1467 * Called with queue lock held. 1468 * Returns the amount of buffer memory freed. 1469 */ 1470static int 1471buf_trim(void) 1472{ 1473 buf_t *bp; 1474 long size; 1475 1476 KASSERT(mutex_owned(&bufcache_lock)); 1477 1478 /* Instruct getnewbuf() to get buffers off the queues */ 1479 if ((bp = getnewbuf(PCATCH, 1, 1)) == NULL) 1480 return 0; 1481 1482 KASSERT((bp->b_cflags & BC_WANTED) == 0); 1483 size = bp->b_bufsize; 1484 bufmem -= size; 1485 if (size > 0) { 1486 buf_mrelease(bp->b_data, size); 1487 bp->b_bcount = bp->b_bufsize = 0; 1488 } 1489 /* brelse() will return the buffer to the global buffer pool */ 1490 brelsel(bp, 0); 1491 return size; 1492} 1493 1494int 1495buf_drain(int n) 1496{ 1497 int size = 0, sz; 1498 1499 KASSERT(mutex_owned(&bufcache_lock)); 1500 1501 while (size < n && bufmem > bufmem_lowater) { 1502 sz = buf_trim(); 1503 if (sz <= 0) 1504 break; 1505 size += sz; 1506 } 1507 1508 return size; 1509} 1510 1511SDT_PROVIDER_DEFINE(io); 1512 1513SDT_PROBE_DEFINE1(io, kernel, , wait__start, "struct buf *"/*bp*/); 1514SDT_PROBE_DEFINE1(io, kernel, , wait__done, "struct buf *"/*bp*/); 1515 1516/* 1517 * Wait for operations on the buffer to complete. 1518 * When they do, extract and return the I/O's error value. 1519 */ 1520int 1521biowait(buf_t *bp) 1522{ 1523 1524 BIOHIST_FUNC(__func__); 1525 1526 KASSERT(ISSET(bp->b_cflags, BC_BUSY)); 1527 KASSERT(bp->b_refcnt > 0); 1528 1529 SDT_PROBE1(io, kernel, , wait__start, bp); 1530 1531 mutex_enter(bp->b_objlock); 1532 1533 BIOHIST_CALLARGS(biohist, "bp=%p, oflags=0x%x", 1534 bp, bp->b_oflags, 0, 0); 1535 1536 while (!ISSET(bp->b_oflags, BO_DONE | BO_DELWRI)) { 1537 BIOHIST_LOG(biohist, "waiting bp=%p", bp, 0, 0, 0); 1538 cv_wait(&bp->b_done, bp->b_objlock); 1539 } 1540 mutex_exit(bp->b_objlock); 1541 1542 SDT_PROBE1(io, kernel, , wait__done, bp); 1543 1544 BIOHIST_LOG(biohist, "return %d", bp->b_error, 0, 0, 0); 1545 1546 return bp->b_error; 1547} 1548 1549/* 1550 * Mark I/O complete on a buffer. 1551 * 1552 * If a callback has been requested, e.g. the pageout 1553 * daemon, do so. Otherwise, awaken waiting processes. 1554 * 1555 * [ Leffler, et al., says on p.247: 1556 * "This routine wakes up the blocked process, frees the buffer 1557 * for an asynchronous write, or, for a request by the pagedaemon 1558 * process, invokes a procedure specified in the buffer structure" ] 1559 * 1560 * In real life, the pagedaemon (or other system processes) wants 1561 * to do async stuff too, and doesn't want the buffer brelse()'d. 1562 * (for swap pager, that puts swap buffers on the free lists (!!!), 1563 * for the vn device, that puts allocated buffers on the free lists!) 1564 */ 1565void 1566biodone(buf_t *bp) 1567{ 1568 int s; 1569 1570 BIOHIST_FUNC(__func__); 1571 1572 KASSERT(!ISSET(bp->b_oflags, BO_DONE)); 1573 1574 if (cpu_intr_p()) { 1575 /* From interrupt mode: defer to a soft interrupt. */ 1576 s = splvm(); 1577 TAILQ_INSERT_TAIL(&curcpu()->ci_data.cpu_biodone, bp, b_actq); 1578 1579 BIOHIST_CALLARGS(biohist, "bp=%p, softint scheduled", 1580 bp, 0, 0, 0); 1581 softint_schedule(biodone_sih); 1582 splx(s); 1583 } else { 1584 /* Process now - the buffer may be freed soon. */ 1585 biodone2(bp); 1586 } 1587} 1588 1589SDT_PROBE_DEFINE1(io, kernel, , done, "struct buf *"/*bp*/); 1590 1591static void 1592biodone2(buf_t *bp) 1593{ 1594 void (*callout)(buf_t *); 1595 1596 SDT_PROBE1(io, kernel, ,done, bp); 1597 1598 BIOHIST_FUNC(__func__); 1599 BIOHIST_CALLARGS(biohist, "bp=%p", bp, 0, 0, 0); 1600 1601 mutex_enter(bp->b_objlock); 1602 /* Note that the transfer is done. */ 1603 if (ISSET(bp->b_oflags, BO_DONE)) 1604 panic("biodone2 already"); 1605 CLR(bp->b_flags, B_COWDONE); 1606 SET(bp->b_oflags, BO_DONE); 1607 BIO_SETPRIO(bp, BPRIO_DEFAULT); 1608 1609 /* Wake up waiting writers. */ 1610 if (!ISSET(bp->b_flags, B_READ)) 1611 vwakeup(bp); 1612 1613 if ((callout = bp->b_iodone) != NULL) { 1614 BIOHIST_LOG(biohist, "callout %p", callout, 0, 0, 0); 1615 1616 /* Note callout done, then call out. */ 1617 KASSERT(!cv_has_waiters(&bp->b_done)); 1618 KERNEL_LOCK(1, NULL); /* XXXSMP */ 1619 bp->b_iodone = NULL; 1620 mutex_exit(bp->b_objlock); 1621 (*callout)(bp); 1622 KERNEL_UNLOCK_ONE(NULL); /* XXXSMP */ 1623 } else if (ISSET(bp->b_flags, B_ASYNC)) { 1624 /* If async, release. */ 1625 BIOHIST_LOG(biohist, "async", 0, 0, 0, 0); 1626 KASSERT(!cv_has_waiters(&bp->b_done)); 1627 mutex_exit(bp->b_objlock); 1628 brelse(bp, 0); 1629 } else { 1630 /* Otherwise just wake up waiters in biowait(). */ 1631 BIOHIST_LOG(biohist, "wake-up", 0, 0, 0, 0); 1632 cv_broadcast(&bp->b_done); 1633 mutex_exit(bp->b_objlock); 1634 } 1635} 1636 1637static void 1638biointr(void *cookie) 1639{ 1640 struct cpu_info *ci; 1641 buf_t *bp; 1642 int s; 1643 1644 BIOHIST_FUNC(__func__); BIOHIST_CALLED(biohist); 1645 1646 ci = curcpu(); 1647 1648 s = splvm(); 1649 while (!TAILQ_EMPTY(&ci->ci_data.cpu_biodone)) { 1650 KASSERT(curcpu() == ci); 1651 1652 bp = TAILQ_FIRST(&ci->ci_data.cpu_biodone); 1653 TAILQ_REMOVE(&ci->ci_data.cpu_biodone, bp, b_actq); 1654 splx(s); 1655 1656 BIOHIST_LOG(biohist, "bp=%p", bp, 0, 0, 0); 1657 biodone2(bp); 1658 1659 s = splvm(); 1660 } 1661 splx(s); 1662} 1663 1664/* 1665 * Wait for all buffers to complete I/O 1666 * Return the number of "stuck" buffers. 1667 */ 1668int 1669buf_syncwait(void) 1670{ 1671 buf_t *bp; 1672 int iter, nbusy, nbusy_prev = 0, ihash; 1673 1674 BIOHIST_FUNC(__func__); BIOHIST_CALLED(biohist); 1675 1676 for (iter = 0; iter < 20;) { 1677 mutex_enter(&bufcache_lock); 1678 nbusy = 0; 1679 for (ihash = 0; ihash < bufhash+1; ihash++) { 1680 LIST_FOREACH(bp, &bufhashtbl[ihash], b_hash) { 1681 if ((bp->b_cflags & (BC_BUSY|BC_INVAL)) == BC_BUSY) 1682 nbusy += ((bp->b_flags & B_READ) == 0); 1683 } 1684 } 1685 mutex_exit(&bufcache_lock); 1686 1687 if (nbusy == 0) 1688 break; 1689 if (nbusy_prev == 0) 1690 nbusy_prev = nbusy; 1691 printf("%d ", nbusy); 1692 kpause("bflush", false, MAX(1, hz / 25 * iter), NULL); 1693 if (nbusy >= nbusy_prev) /* we didn't flush anything */ 1694 iter++; 1695 else 1696 nbusy_prev = nbusy; 1697 } 1698 1699 if (nbusy) { 1700#if defined(DEBUG) || defined(DEBUG_HALT_BUSY) 1701 printf("giving up\nPrinting vnodes for busy buffers\n"); 1702 for (ihash = 0; ihash < bufhash+1; ihash++) { 1703 LIST_FOREACH(bp, &bufhashtbl[ihash], b_hash) { 1704 if ((bp->b_cflags & (BC_BUSY|BC_INVAL)) == BC_BUSY && 1705 (bp->b_flags & B_READ) == 0) 1706 vprint(NULL, bp->b_vp); 1707 } 1708 } 1709#endif 1710 } 1711 1712 return nbusy; 1713} 1714 1715static void 1716sysctl_fillbuf(buf_t *i, struct buf_sysctl *o) 1717{ 1718 1719 o->b_flags = i->b_flags | i->b_cflags | i->b_oflags; 1720 o->b_error = i->b_error; 1721 o->b_prio = i->b_prio; 1722 o->b_dev = i->b_dev; 1723 o->b_bufsize = i->b_bufsize; 1724 o->b_bcount = i->b_bcount; 1725 o->b_resid = i->b_resid; 1726 o->b_addr = PTRTOUINT64(i->b_data); 1727 o->b_blkno = i->b_blkno; 1728 o->b_rawblkno = i->b_rawblkno; 1729 o->b_iodone = PTRTOUINT64(i->b_iodone); 1730 o->b_proc = PTRTOUINT64(i->b_proc); 1731 o->b_vp = PTRTOUINT64(i->b_vp); 1732 o->b_saveaddr = PTRTOUINT64(i->b_saveaddr); 1733 o->b_lblkno = i->b_lblkno; 1734} 1735 1736#define KERN_BUFSLOP 20 1737static int 1738sysctl_dobuf(SYSCTLFN_ARGS) 1739{ 1740 buf_t *bp; 1741 struct buf_sysctl bs; 1742 struct bqueue *bq; 1743 char *dp; 1744 u_int i, op, arg; 1745 size_t len, needed, elem_size, out_size; 1746 int error, elem_count, retries; 1747 1748 if (namelen == 1 && name[0] == CTL_QUERY) 1749 return (sysctl_query(SYSCTLFN_CALL(rnode))); 1750 1751 if (namelen != 4) 1752 return (EINVAL); 1753 1754 retries = 100; 1755 retry: 1756 dp = oldp; 1757 len = (oldp != NULL) ? *oldlenp : 0; 1758 op = name[0]; 1759 arg = name[1]; 1760 elem_size = name[2]; 1761 elem_count = name[3]; 1762 out_size = MIN(sizeof(bs), elem_size); 1763 1764 /* 1765 * at the moment, these are just "placeholders" to make the 1766 * API for retrieving kern.buf data more extensible in the 1767 * future. 1768 * 1769 * XXX kern.buf currently has "netbsd32" issues. hopefully 1770 * these will be resolved at a later point. 1771 */ 1772 if (op != KERN_BUF_ALL || arg != KERN_BUF_ALL || 1773 elem_size < 1 || elem_count < 0) 1774 return (EINVAL); 1775 1776 error = 0; 1777 needed = 0; 1778 sysctl_unlock(); 1779 mutex_enter(&bufcache_lock); 1780 for (i = 0; i < BQUEUES; i++) { 1781 bq = &bufqueues[i]; 1782 TAILQ_FOREACH(bp, &bq->bq_queue, b_freelist) { 1783 bq->bq_marker = bp; 1784 if (len >= elem_size && elem_count > 0) { 1785 sysctl_fillbuf(bp, &bs); 1786 mutex_exit(&bufcache_lock); 1787 error = copyout(&bs, dp, out_size); 1788 mutex_enter(&bufcache_lock); 1789 if (error) 1790 break; 1791 if (bq->bq_marker != bp) { 1792 /* 1793 * This sysctl node is only for 1794 * statistics. Retry; if the 1795 * queue keeps changing, then 1796 * bail out. 1797 */ 1798 if (retries-- == 0) { 1799 error = EAGAIN; 1800 break; 1801 } 1802 mutex_exit(&bufcache_lock); 1803 sysctl_relock(); 1804 goto retry; 1805 } 1806 dp += elem_size; 1807 len -= elem_size; 1808 } 1809 needed += elem_size; 1810 if (elem_count > 0 && elem_count != INT_MAX) 1811 elem_count--; 1812 } 1813 if (error != 0) 1814 break; 1815 } 1816 mutex_exit(&bufcache_lock); 1817 sysctl_relock(); 1818 1819 *oldlenp = needed; 1820 if (oldp == NULL) 1821 *oldlenp += KERN_BUFSLOP * sizeof(buf_t); 1822 1823 return (error); 1824} 1825 1826static int 1827sysctl_bufvm_update(SYSCTLFN_ARGS) 1828{ 1829 int error, rv; 1830 struct sysctlnode node; 1831 unsigned int temp_bufcache; 1832 unsigned long temp_water; 1833 1834 /* Take a copy of the supplied node and its data */ 1835 node = *rnode; 1836 if (node.sysctl_data == &bufcache) { 1837 node.sysctl_data = &temp_bufcache; 1838 temp_bufcache = *(unsigned int *)rnode->sysctl_data; 1839 } else { 1840 node.sysctl_data = &temp_water; 1841 temp_water = *(unsigned long *)rnode->sysctl_data; 1842 } 1843 1844 /* Update the copy */ 1845 error = sysctl_lookup(SYSCTLFN_CALL(&node)); 1846 if (error || newp == NULL) 1847 return (error); 1848 1849 if (rnode->sysctl_data == &bufcache) { 1850 if (temp_bufcache > 100) 1851 return (EINVAL); 1852 bufcache = temp_bufcache; 1853 buf_setwm(); 1854 } else if (rnode->sysctl_data == &bufmem_lowater) { 1855 if (bufmem_hiwater - temp_water < 16) 1856 return (EINVAL); 1857 bufmem_lowater = temp_water; 1858 } else if (rnode->sysctl_data == &bufmem_hiwater) { 1859 if (temp_water - bufmem_lowater < 16) 1860 return (EINVAL); 1861 bufmem_hiwater = temp_water; 1862 } else 1863 return (EINVAL); 1864 1865 /* Drain until below new high water mark */ 1866 sysctl_unlock(); 1867 mutex_enter(&bufcache_lock); 1868 while (bufmem > bufmem_hiwater) { 1869 rv = buf_drain((bufmem - bufmem_hiwater) / (2 * 1024)); 1870 if (rv <= 0) 1871 break; 1872 } 1873 mutex_exit(&bufcache_lock); 1874 sysctl_relock(); 1875 1876 return 0; 1877} 1878 1879static struct sysctllog *vfsbio_sysctllog; 1880 1881static void 1882sysctl_kern_buf_setup(void) 1883{ 1884 1885 sysctl_createv(&vfsbio_sysctllog, 0, NULL, NULL, 1886 CTLFLAG_PERMANENT, 1887 CTLTYPE_NODE, "buf", 1888 SYSCTL_DESCR("Kernel buffer cache information"), 1889 sysctl_dobuf, 0, NULL, 0, 1890 CTL_KERN, KERN_BUF, CTL_EOL); 1891} 1892 1893static void 1894sysctl_vm_buf_setup(void) 1895{ 1896 1897 sysctl_createv(&vfsbio_sysctllog, 0, NULL, NULL, 1898 CTLFLAG_PERMANENT|CTLFLAG_READWRITE, 1899 CTLTYPE_INT, "bufcache", 1900 SYSCTL_DESCR("Percentage of physical memory to use for " 1901 "buffer cache"), 1902 sysctl_bufvm_update, 0, &bufcache, 0, 1903 CTL_VM, CTL_CREATE, CTL_EOL); 1904 sysctl_createv(&vfsbio_sysctllog, 0, NULL, NULL, 1905 CTLFLAG_PERMANENT|CTLFLAG_READONLY, 1906 CTLTYPE_LONG, "bufmem", 1907 SYSCTL_DESCR("Amount of kernel memory used by buffer " 1908 "cache"), 1909 NULL, 0, &bufmem, 0, 1910 CTL_VM, CTL_CREATE, CTL_EOL); 1911 sysctl_createv(&vfsbio_sysctllog, 0, NULL, NULL, 1912 CTLFLAG_PERMANENT|CTLFLAG_READWRITE, 1913 CTLTYPE_LONG, "bufmem_lowater", 1914 SYSCTL_DESCR("Minimum amount of kernel memory to " 1915 "reserve for buffer cache"), 1916 sysctl_bufvm_update, 0, &bufmem_lowater, 0, 1917 CTL_VM, CTL_CREATE, CTL_EOL); 1918 sysctl_createv(&vfsbio_sysctllog, 0, NULL, NULL, 1919 CTLFLAG_PERMANENT|CTLFLAG_READWRITE, 1920 CTLTYPE_LONG, "bufmem_hiwater", 1921 SYSCTL_DESCR("Maximum amount of kernel memory to use " 1922 "for buffer cache"), 1923 sysctl_bufvm_update, 0, &bufmem_hiwater, 0, 1924 CTL_VM, CTL_CREATE, CTL_EOL); 1925} 1926 1927#ifdef DEBUG 1928/* 1929 * Print out statistics on the current allocation of the buffer pool. 1930 * Can be enabled to print out on every ``sync'' by setting "syncprt" 1931 * in vfs_syscalls.c using sysctl. 1932 */ 1933void 1934vfs_bufstats(void) 1935{ 1936 int i, j, count; 1937 buf_t *bp; 1938 struct bqueue *dp; 1939 int counts[MAXBSIZE / MIN_PAGE_SIZE + 1]; 1940 static const char *bname[BQUEUES] = { "LOCKED", "LRU", "AGE" }; 1941 1942 for (dp = bufqueues, i = 0; dp < &bufqueues[BQUEUES]; dp++, i++) { 1943 count = 0; 1944 memset(counts, 0, sizeof(counts)); 1945 TAILQ_FOREACH(bp, &dp->bq_queue, b_freelist) { 1946 counts[bp->b_bufsize / PAGE_SIZE]++; 1947 count++; 1948 } 1949 printf("%s: total-%d", bname[i], count); 1950 for (j = 0; j <= MAXBSIZE / PAGE_SIZE; j++) 1951 if (counts[j] != 0) 1952 printf(", %d-%d", j * PAGE_SIZE, counts[j]); 1953 printf("\n"); 1954 } 1955} 1956#endif /* DEBUG */ 1957 1958/* ------------------------------ */ 1959 1960buf_t * 1961getiobuf(struct vnode *vp, bool waitok) 1962{ 1963 buf_t *bp; 1964 1965 bp = pool_cache_get(bufio_cache, (waitok ? PR_WAITOK : PR_NOWAIT)); 1966 if (bp == NULL) 1967 return bp; 1968 1969 buf_init(bp); 1970 1971 if ((bp->b_vp = vp) != NULL) { 1972 bp->b_objlock = vp->v_interlock; 1973 } else { 1974 KASSERT(bp->b_objlock == &buffer_lock); 1975 } 1976 1977 return bp; 1978} 1979 1980void 1981putiobuf(buf_t *bp) 1982{ 1983 1984 buf_destroy(bp); 1985 pool_cache_put(bufio_cache, bp); 1986} 1987 1988/* 1989 * nestiobuf_iodone: b_iodone callback for nested buffers. 1990 */ 1991 1992void 1993nestiobuf_iodone(buf_t *bp) 1994{ 1995 buf_t *mbp = bp->b_private; 1996 int error; 1997 int donebytes; 1998 1999 KASSERT(bp->b_bcount <= bp->b_bufsize); 2000 KASSERT(mbp != bp); 2001 2002 error = bp->b_error; 2003 if (bp->b_error == 0 && 2004 (bp->b_bcount < bp->b_bufsize || bp->b_resid > 0)) { 2005 /* 2006 * Not all got transfered, raise an error. We have no way to 2007 * propagate these conditions to mbp. 2008 */ 2009 error = EIO; 2010 } 2011 2012 donebytes = bp->b_bufsize; 2013 2014 putiobuf(bp); 2015 nestiobuf_done(mbp, donebytes, error); 2016} 2017 2018/* 2019 * nestiobuf_setup: setup a "nested" buffer. 2020 * 2021 * => 'mbp' is a "master" buffer which is being divided into sub pieces. 2022 * => 'bp' should be a buffer allocated by getiobuf. 2023 * => 'offset' is a byte offset in the master buffer. 2024 * => 'size' is a size in bytes of this nested buffer. 2025 */ 2026 2027void 2028nestiobuf_setup(buf_t *mbp, buf_t *bp, int offset, size_t size) 2029{ 2030 const int b_read = mbp->b_flags & B_READ; 2031 struct vnode *vp = mbp->b_vp; 2032 2033 KASSERT(mbp->b_bcount >= offset + size); 2034 bp->b_vp = vp; 2035 bp->b_dev = mbp->b_dev; 2036 bp->b_objlock = mbp->b_objlock; 2037 bp->b_cflags = BC_BUSY; 2038 bp->b_flags = B_ASYNC | b_read; 2039 bp->b_iodone = nestiobuf_iodone; 2040 bp->b_data = (char *)mbp->b_data + offset; 2041 bp->b_resid = bp->b_bcount = size; 2042 bp->b_bufsize = bp->b_bcount; 2043 bp->b_private = mbp; 2044 BIO_COPYPRIO(bp, mbp); 2045 if (!b_read && vp != NULL) { 2046 mutex_enter(vp->v_interlock); 2047 vp->v_numoutput++; 2048 mutex_exit(vp->v_interlock); 2049 } 2050} 2051 2052/* 2053 * nestiobuf_done: propagate completion to the master buffer. 2054 * 2055 * => 'donebytes' specifies how many bytes in the 'mbp' is completed. 2056 * => 'error' is an errno(2) that 'donebytes' has been completed with. 2057 */ 2058 2059void 2060nestiobuf_done(buf_t *mbp, int donebytes, int error) 2061{ 2062 2063 if (donebytes == 0) { 2064 return; 2065 } 2066 mutex_enter(mbp->b_objlock); 2067 KASSERT(mbp->b_resid >= donebytes); 2068 mbp->b_resid -= donebytes; 2069 if (error) 2070 mbp->b_error = error; 2071 if (mbp->b_resid == 0) { 2072 if (mbp->b_error) 2073 mbp->b_resid = mbp->b_bcount; 2074 mutex_exit(mbp->b_objlock); 2075 biodone(mbp); 2076 } else 2077 mutex_exit(mbp->b_objlock); 2078} 2079 2080void 2081buf_init(buf_t *bp) 2082{ 2083 2084 cv_init(&bp->b_busy, "biolock"); 2085 cv_init(&bp->b_done, "biowait"); 2086 bp->b_dev = NODEV; 2087 bp->b_error = 0; 2088 bp->b_flags = 0; 2089 bp->b_cflags = 0; 2090 bp->b_oflags = 0; 2091 bp->b_objlock = &buffer_lock; 2092 bp->b_iodone = NULL; 2093 bp->b_refcnt = 1; 2094 bp->b_dev = NODEV; 2095 bp->b_vnbufs.le_next = NOLIST; 2096 BIO_SETPRIO(bp, BPRIO_DEFAULT); 2097} 2098 2099void 2100buf_destroy(buf_t *bp) 2101{ 2102 2103 cv_destroy(&bp->b_done); 2104 cv_destroy(&bp->b_busy); 2105} 2106 2107int 2108bbusy(buf_t *bp, bool intr, int timo, kmutex_t *interlock) 2109{ 2110 int error; 2111 2112 KASSERT(mutex_owned(&bufcache_lock)); 2113 2114 if ((bp->b_cflags & BC_BUSY) != 0) { 2115 if (curlwp == uvm.pagedaemon_lwp) 2116 return EDEADLK; 2117 bp->b_cflags |= BC_WANTED; 2118 bref(bp); 2119 if (interlock != NULL) 2120 mutex_exit(interlock); 2121 if (intr) { 2122 error = cv_timedwait_sig(&bp->b_busy, &bufcache_lock, 2123 timo); 2124 } else { 2125 error = cv_timedwait(&bp->b_busy, &bufcache_lock, 2126 timo); 2127 } 2128 brele(bp); 2129 if (interlock != NULL) 2130 mutex_enter(interlock); 2131 if (error != 0) 2132 return error; 2133 return EPASSTHROUGH; 2134 } 2135 bp->b_cflags |= BC_BUSY; 2136 2137 return 0; 2138} 2139