1 /* $NetBSD: vfs_bio.c,v 1.236 2012/02/01 23:43:49 para 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 1.236 2012/02/01 23:43:49 para Exp $"); 127 128 #include "opt_bufcache.h" 129 130 #include <sys/param.h> 131 #include <sys/systm.h> 132 #include <sys/kernel.h> 133 #include <sys/proc.h> 134 #include <sys/buf.h> 135 #include <sys/vnode.h> 136 #include <sys/mount.h> 137 #include <sys/resourcevar.h> 138 #include <sys/sysctl.h> 139 #include <sys/conf.h> 140 #include <sys/kauth.h> 141 #include <sys/fstrans.h> 142 #include <sys/intr.h> 143 #include <sys/cpu.h> 144 #include <sys/wapbl.h> 145 #include <sys/bitops.h> 146 147 #include <uvm/uvm.h> /* extern struct uvm uvm */ 148 149 #include <miscfs/specfs/specdev.h> 150 151 #ifndef BUFPAGES 152 # define BUFPAGES 0 153 #endif 154 155 #ifdef BUFCACHE 156 # if (BUFCACHE < 5) || (BUFCACHE > 95) 157 # error BUFCACHE is not between 5 and 95 158 # endif 159 #else 160 # define BUFCACHE 15 161 #endif 162 163 u_int nbuf; /* desired number of buffer headers */ 164 u_int bufpages = BUFPAGES; /* optional hardwired count */ 165 u_int bufcache = BUFCACHE; /* max % of RAM to use for buffer cache */ 166 167 /* Function prototypes */ 168 struct bqueue; 169 170 static void buf_setwm(void); 171 static int buf_trim(void); 172 static void *bufpool_page_alloc(struct pool *, int); 173 static void bufpool_page_free(struct pool *, void *); 174 static buf_t *bio_doread(struct vnode *, daddr_t, int, 175 kauth_cred_t, int); 176 static buf_t *getnewbuf(int, int, int); 177 static int buf_lotsfree(void); 178 static int buf_canrelease(void); 179 static u_long buf_mempoolidx(u_long); 180 static u_long buf_roundsize(u_long); 181 static void *buf_alloc(size_t); 182 static void buf_mrelease(void *, size_t); 183 static void binsheadfree(buf_t *, struct bqueue *); 184 static void binstailfree(buf_t *, struct bqueue *); 185 #ifdef DEBUG 186 static int checkfreelist(buf_t *, struct bqueue *, int); 187 #endif 188 static void biointr(void *); 189 static void biodone2(buf_t *); 190 static void bref(buf_t *); 191 static void brele(buf_t *); 192 static void sysctl_kern_buf_setup(void); 193 static void sysctl_vm_buf_setup(void); 194 195 /* 196 * Definitions for the buffer hash lists. 197 */ 198 #define BUFHASH(dvp, lbn) \ 199 (&bufhashtbl[(((long)(dvp) >> 8) + (int)(lbn)) & bufhash]) 200 LIST_HEAD(bufhashhdr, buf) *bufhashtbl, invalhash; 201 u_long bufhash; 202 struct bqueue bufqueues[BQUEUES]; 203 204 static kcondvar_t needbuffer_cv; 205 206 /* 207 * Buffer queue lock. 208 */ 209 kmutex_t bufcache_lock; 210 kmutex_t buffer_lock; 211 212 /* Software ISR for completed transfers. */ 213 static void *biodone_sih; 214 215 /* Buffer pool for I/O buffers. */ 216 static pool_cache_t buf_cache; 217 static pool_cache_t bufio_cache; 218 219 #define MEMPOOL_INDEX_OFFSET (ilog2(DEV_BSIZE)) /* smallest pool is 512 bytes */ 220 #define NMEMPOOLS (ilog2(MAXBSIZE) - MEMPOOL_INDEX_OFFSET + 1) 221 __CTASSERT((1 << (NMEMPOOLS + MEMPOOL_INDEX_OFFSET - 1)) == MAXBSIZE); 222 223 /* Buffer memory pools */ 224 static struct pool bmempools[NMEMPOOLS]; 225 226 static struct vm_map *buf_map; 227 228 /* 229 * Buffer memory pool allocator. 230 */ 231 static void * 232 bufpool_page_alloc(struct pool *pp, int flags) 233 { 234 235 return (void *)uvm_km_alloc(buf_map, 236 MAXBSIZE, MAXBSIZE, 237 ((flags & PR_WAITOK) ? 0 : UVM_KMF_NOWAIT|UVM_KMF_TRYLOCK) 238 | UVM_KMF_WIRED); 239 } 240 241 static void 242 bufpool_page_free(struct pool *pp, void *v) 243 { 244 245 uvm_km_free(buf_map, (vaddr_t)v, MAXBSIZE, UVM_KMF_WIRED); 246 } 247 248 static struct pool_allocator bufmempool_allocator = { 249 .pa_alloc = bufpool_page_alloc, 250 .pa_free = bufpool_page_free, 251 .pa_pagesz = MAXBSIZE, 252 }; 253 254 /* Buffer memory management variables */ 255 u_long bufmem_valimit; 256 u_long bufmem_hiwater; 257 u_long bufmem_lowater; 258 u_long bufmem; 259 260 /* 261 * MD code can call this to set a hard limit on the amount 262 * of virtual memory used by the buffer cache. 263 */ 264 int 265 buf_setvalimit(vsize_t sz) 266 { 267 268 /* We need to accommodate at least NMEMPOOLS of MAXBSIZE each */ 269 if (sz < NMEMPOOLS * MAXBSIZE) 270 return EINVAL; 271 272 bufmem_valimit = sz; 273 return 0; 274 } 275 276 static void 277 buf_setwm(void) 278 { 279 280 bufmem_hiwater = buf_memcalc(); 281 /* lowater is approx. 2% of memory (with bufcache = 15) */ 282 #define BUFMEM_WMSHIFT 3 283 #define BUFMEM_HIWMMIN (64 * 1024 << BUFMEM_WMSHIFT) 284 if (bufmem_hiwater < BUFMEM_HIWMMIN) 285 /* Ensure a reasonable minimum value */ 286 bufmem_hiwater = BUFMEM_HIWMMIN; 287 bufmem_lowater = bufmem_hiwater >> BUFMEM_WMSHIFT; 288 } 289 290 #ifdef DEBUG 291 int debug_verify_freelist = 0; 292 static int 293 checkfreelist(buf_t *bp, struct bqueue *dp, int ison) 294 { 295 buf_t *b; 296 297 if (!debug_verify_freelist) 298 return 1; 299 300 TAILQ_FOREACH(b, &dp->bq_queue, b_freelist) { 301 if (b == bp) 302 return ison ? 1 : 0; 303 } 304 305 return ison ? 0 : 1; 306 } 307 #endif 308 309 /* 310 * Insq/Remq for the buffer hash lists. 311 * Call with buffer queue locked. 312 */ 313 static void 314 binsheadfree(buf_t *bp, struct bqueue *dp) 315 { 316 317 KASSERT(mutex_owned(&bufcache_lock)); 318 KASSERT(bp->b_freelistindex == -1); 319 TAILQ_INSERT_HEAD(&dp->bq_queue, bp, b_freelist); 320 dp->bq_bytes += bp->b_bufsize; 321 bp->b_freelistindex = dp - bufqueues; 322 } 323 324 static void 325 binstailfree(buf_t *bp, struct bqueue *dp) 326 { 327 328 KASSERT(mutex_owned(&bufcache_lock)); 329 KASSERT(bp->b_freelistindex == -1); 330 TAILQ_INSERT_TAIL(&dp->bq_queue, bp, b_freelist); 331 dp->bq_bytes += bp->b_bufsize; 332 bp->b_freelistindex = dp - bufqueues; 333 } 334 335 void 336 bremfree(buf_t *bp) 337 { 338 struct bqueue *dp; 339 int bqidx = bp->b_freelistindex; 340 341 KASSERT(mutex_owned(&bufcache_lock)); 342 343 KASSERT(bqidx != -1); 344 dp = &bufqueues[bqidx]; 345 KDASSERT(checkfreelist(bp, dp, 1)); 346 KASSERT(dp->bq_bytes >= bp->b_bufsize); 347 TAILQ_REMOVE(&dp->bq_queue, bp, b_freelist); 348 dp->bq_bytes -= bp->b_bufsize; 349 350 /* For the sysctl helper. */ 351 if (bp == dp->bq_marker) 352 dp->bq_marker = NULL; 353 354 #if defined(DIAGNOSTIC) 355 bp->b_freelistindex = -1; 356 #endif /* defined(DIAGNOSTIC) */ 357 } 358 359 /* 360 * Add a reference to an buffer structure that came from buf_cache. 361 */ 362 static inline void 363 bref(buf_t *bp) 364 { 365 366 KASSERT(mutex_owned(&bufcache_lock)); 367 KASSERT(bp->b_refcnt > 0); 368 369 bp->b_refcnt++; 370 } 371 372 /* 373 * Free an unused buffer structure that came from buf_cache. 374 */ 375 static inline void 376 brele(buf_t *bp) 377 { 378 379 KASSERT(mutex_owned(&bufcache_lock)); 380 KASSERT(bp->b_refcnt > 0); 381 382 if (bp->b_refcnt-- == 1) { 383 buf_destroy(bp); 384 #ifdef DEBUG 385 memset((char *)bp, 0, sizeof(*bp)); 386 #endif 387 pool_cache_put(buf_cache, bp); 388 } 389 } 390 391 /* 392 * note that for some ports this is used by pmap bootstrap code to 393 * determine kva size. 394 */ 395 u_long 396 buf_memcalc(void) 397 { 398 u_long n; 399 400 /* 401 * Determine the upper bound of memory to use for buffers. 402 * 403 * - If bufpages is specified, use that as the number 404 * pages. 405 * 406 * - Otherwise, use bufcache as the percentage of 407 * physical memory. 408 */ 409 if (bufpages != 0) { 410 n = bufpages; 411 } else { 412 if (bufcache < 5) { 413 printf("forcing bufcache %d -> 5", bufcache); 414 bufcache = 5; 415 } 416 if (bufcache > 95) { 417 printf("forcing bufcache %d -> 95", bufcache); 418 bufcache = 95; 419 } 420 n = calc_cache_size(buf_map, bufcache, 421 (buf_map != kernel_map) ? 100 : BUFCACHE_VA_MAXPCT) 422 / PAGE_SIZE; 423 } 424 425 n <<= PAGE_SHIFT; 426 if (bufmem_valimit != 0 && n > bufmem_valimit) 427 n = bufmem_valimit; 428 429 return (n); 430 } 431 432 /* 433 * Initialize buffers and hash links for buffers. 434 */ 435 void 436 bufinit(void) 437 { 438 struct bqueue *dp; 439 int use_std; 440 u_int i; 441 442 mutex_init(&bufcache_lock, MUTEX_DEFAULT, IPL_NONE); 443 mutex_init(&buffer_lock, MUTEX_DEFAULT, IPL_NONE); 444 cv_init(&needbuffer_cv, "needbuf"); 445 446 if (bufmem_valimit != 0) { 447 vaddr_t minaddr = 0, maxaddr; 448 buf_map = uvm_km_suballoc(kernel_map, &minaddr, &maxaddr, 449 bufmem_valimit, 0, false, 0); 450 if (buf_map == NULL) 451 panic("bufinit: cannot allocate submap"); 452 } else 453 buf_map = kernel_map; 454 455 /* 456 * Initialize buffer cache memory parameters. 457 */ 458 bufmem = 0; 459 buf_setwm(); 460 461 /* On "small" machines use small pool page sizes where possible */ 462 use_std = (physmem < atop(16*1024*1024)); 463 464 /* 465 * Also use them on systems that can map the pool pages using 466 * a direct-mapped segment. 467 */ 468 #ifdef PMAP_MAP_POOLPAGE 469 use_std = 1; 470 #endif 471 472 buf_cache = pool_cache_init(sizeof(buf_t), 0, 0, 0, 473 "bufpl", NULL, IPL_SOFTBIO, NULL, NULL, NULL); 474 bufio_cache = pool_cache_init(sizeof(buf_t), 0, 0, 0, 475 "biopl", NULL, IPL_BIO, NULL, NULL, NULL); 476 477 for (i = 0; i < NMEMPOOLS; i++) { 478 struct pool_allocator *pa; 479 struct pool *pp = &bmempools[i]; 480 u_int size = 1 << (i + MEMPOOL_INDEX_OFFSET); 481 char *name = kmem_alloc(8, KM_SLEEP); /* XXX: never freed */ 482 if (__predict_false(size >= 1048576)) 483 (void)snprintf(name, 8, "buf%um", size / 1048576); 484 else if (__predict_true(size >= 1024)) 485 (void)snprintf(name, 8, "buf%uk", size / 1024); 486 else 487 (void)snprintf(name, 8, "buf%ub", size); 488 pa = (size <= PAGE_SIZE && use_std) 489 ? &pool_allocator_nointr 490 : &bufmempool_allocator; 491 pool_init(pp, size, 0, 0, 0, name, pa, IPL_NONE); 492 pool_setlowat(pp, 1); 493 pool_sethiwat(pp, 1); 494 } 495 496 /* Initialize the buffer queues */ 497 for (dp = bufqueues; dp < &bufqueues[BQUEUES]; dp++) { 498 TAILQ_INIT(&dp->bq_queue); 499 dp->bq_bytes = 0; 500 } 501 502 /* 503 * Estimate hash table size based on the amount of memory we 504 * intend to use for the buffer cache. The average buffer 505 * size is dependent on our clients (i.e. filesystems). 506 * 507 * For now, use an empirical 3K per buffer. 508 */ 509 nbuf = (bufmem_hiwater / 1024) / 3; 510 bufhashtbl = hashinit(nbuf, HASH_LIST, true, &bufhash); 511 512 sysctl_kern_buf_setup(); 513 sysctl_vm_buf_setup(); 514 } 515 516 void 517 bufinit2(void) 518 { 519 520 biodone_sih = softint_establish(SOFTINT_BIO | SOFTINT_MPSAFE, biointr, 521 NULL); 522 if (biodone_sih == NULL) 523 panic("bufinit2: can't establish soft interrupt"); 524 } 525 526 static int 527 buf_lotsfree(void) 528 { 529 int try, thresh; 530 531 /* Always allocate if less than the low water mark. */ 532 if (bufmem < bufmem_lowater) 533 return 1; 534 535 /* Never allocate if greater than the high water mark. */ 536 if (bufmem > bufmem_hiwater) 537 return 0; 538 539 /* If there's anything on the AGE list, it should be eaten. */ 540 if (TAILQ_FIRST(&bufqueues[BQ_AGE].bq_queue) != NULL) 541 return 0; 542 543 /* 544 * The probabily of getting a new allocation is inversely 545 * proportional to the current size of the cache, using 546 * a granularity of 16 steps. 547 */ 548 try = random() & 0x0000000fL; 549 550 /* Don't use "16 * bufmem" here to avoid a 32-bit overflow. */ 551 thresh = (bufmem - bufmem_lowater) / 552 ((bufmem_hiwater - bufmem_lowater) / 16); 553 554 if (try >= thresh) 555 return 1; 556 557 /* Otherwise don't allocate. */ 558 return 0; 559 } 560 561 /* 562 * Return estimate of bytes we think need to be 563 * released to help resolve low memory conditions. 564 * 565 * => called with bufcache_lock held. 566 */ 567 static int 568 buf_canrelease(void) 569 { 570 int pagedemand, ninvalid = 0; 571 572 KASSERT(mutex_owned(&bufcache_lock)); 573 574 if (bufmem < bufmem_lowater) 575 return 0; 576 577 if (bufmem > bufmem_hiwater) 578 return bufmem - bufmem_hiwater; 579 580 ninvalid += bufqueues[BQ_AGE].bq_bytes; 581 582 pagedemand = uvmexp.freetarg - uvmexp.free; 583 if (pagedemand < 0) 584 return ninvalid; 585 return MAX(ninvalid, MIN(2 * MAXBSIZE, 586 MIN((bufmem - bufmem_lowater) / 16, pagedemand * PAGE_SIZE))); 587 } 588 589 /* 590 * Buffer memory allocation helper functions 591 */ 592 static u_long 593 buf_mempoolidx(u_long size) 594 { 595 u_int n = 0; 596 597 size -= 1; 598 size >>= MEMPOOL_INDEX_OFFSET; 599 while (size) { 600 size >>= 1; 601 n += 1; 602 } 603 if (n >= NMEMPOOLS) 604 panic("buf mem pool index %d", n); 605 return n; 606 } 607 608 static u_long 609 buf_roundsize(u_long size) 610 { 611 /* Round up to nearest power of 2 */ 612 return (1 << (buf_mempoolidx(size) + MEMPOOL_INDEX_OFFSET)); 613 } 614 615 static void * 616 buf_alloc(size_t size) 617 { 618 u_int n = buf_mempoolidx(size); 619 void *addr; 620 621 while (1) { 622 addr = pool_get(&bmempools[n], PR_NOWAIT); 623 if (addr != NULL) 624 break; 625 626 /* No memory, see if we can free some. If so, try again */ 627 mutex_enter(&bufcache_lock); 628 if (buf_drain(1) > 0) { 629 mutex_exit(&bufcache_lock); 630 continue; 631 } 632 633 if (curlwp == uvm.pagedaemon_lwp) { 634 mutex_exit(&bufcache_lock); 635 return NULL; 636 } 637 638 /* Wait for buffers to arrive on the LRU queue */ 639 cv_timedwait(&needbuffer_cv, &bufcache_lock, hz / 4); 640 mutex_exit(&bufcache_lock); 641 } 642 643 return addr; 644 } 645 646 static void 647 buf_mrelease(void *addr, size_t size) 648 { 649 650 pool_put(&bmempools[buf_mempoolidx(size)], addr); 651 } 652 653 /* 654 * bread()/breadn() helper. 655 */ 656 static buf_t * 657 bio_doread(struct vnode *vp, daddr_t blkno, int size, kauth_cred_t cred, 658 int async) 659 { 660 buf_t *bp; 661 struct mount *mp; 662 663 bp = getblk(vp, blkno, size, 0, 0); 664 665 #ifdef DIAGNOSTIC 666 if (bp == NULL) { 667 panic("bio_doread: no such buf"); 668 } 669 #endif 670 671 /* 672 * If buffer does not have data valid, start a read. 673 * Note that if buffer is BC_INVAL, getblk() won't return it. 674 * Therefore, it's valid if its I/O has completed or been delayed. 675 */ 676 if (!ISSET(bp->b_oflags, (BO_DONE | BO_DELWRI))) { 677 /* Start I/O for the buffer. */ 678 SET(bp->b_flags, B_READ | async); 679 if (async) 680 BIO_SETPRIO(bp, BPRIO_TIMELIMITED); 681 else 682 BIO_SETPRIO(bp, BPRIO_TIMECRITICAL); 683 VOP_STRATEGY(vp, bp); 684 685 /* Pay for the read. */ 686 curlwp->l_ru.ru_inblock++; 687 } else if (async) 688 brelse(bp, 0); 689 690 if (vp->v_type == VBLK) 691 mp = vp->v_specmountpoint; 692 else 693 mp = vp->v_mount; 694 695 /* 696 * Collect statistics on synchronous and asynchronous reads. 697 * Reads from block devices are charged to their associated 698 * filesystem (if any). 699 */ 700 if (mp != NULL) { 701 if (async == 0) 702 mp->mnt_stat.f_syncreads++; 703 else 704 mp->mnt_stat.f_asyncreads++; 705 } 706 707 return (bp); 708 } 709 710 /* 711 * Read a disk block. 712 * This algorithm described in Bach (p.54). 713 */ 714 int 715 bread(struct vnode *vp, daddr_t blkno, int size, kauth_cred_t cred, 716 int flags, buf_t **bpp) 717 { 718 buf_t *bp; 719 int error; 720 721 /* Get buffer for block. */ 722 bp = *bpp = bio_doread(vp, blkno, size, cred, 0); 723 724 /* Wait for the read to complete, and return result. */ 725 error = biowait(bp); 726 if (error == 0 && (flags & B_MODIFY) != 0) 727 error = fscow_run(bp, true); 728 729 return error; 730 } 731 732 /* 733 * Read-ahead multiple disk blocks. The first is sync, the rest async. 734 * Trivial modification to the breada algorithm presented in Bach (p.55). 735 */ 736 int 737 breadn(struct vnode *vp, daddr_t blkno, int size, daddr_t *rablks, 738 int *rasizes, int nrablks, kauth_cred_t cred, int flags, buf_t **bpp) 739 { 740 buf_t *bp; 741 int error, i; 742 743 bp = *bpp = bio_doread(vp, blkno, size, cred, 0); 744 745 /* 746 * For each of the read-ahead blocks, start a read, if necessary. 747 */ 748 mutex_enter(&bufcache_lock); 749 for (i = 0; i < nrablks; i++) { 750 /* If it's in the cache, just go on to next one. */ 751 if (incore(vp, rablks[i])) 752 continue; 753 754 /* Get a buffer for the read-ahead block */ 755 mutex_exit(&bufcache_lock); 756 (void) bio_doread(vp, rablks[i], rasizes[i], cred, B_ASYNC); 757 mutex_enter(&bufcache_lock); 758 } 759 mutex_exit(&bufcache_lock); 760 761 /* Otherwise, we had to start a read for it; wait until it's valid. */ 762 error = biowait(bp); 763 if (error == 0 && (flags & B_MODIFY) != 0) 764 error = fscow_run(bp, true); 765 return error; 766 } 767 768 /* 769 * Block write. Described in Bach (p.56) 770 */ 771 int 772 bwrite(buf_t *bp) 773 { 774 int rv, sync, wasdelayed; 775 struct vnode *vp; 776 struct mount *mp; 777 778 KASSERT(ISSET(bp->b_cflags, BC_BUSY)); 779 KASSERT(!cv_has_waiters(&bp->b_done)); 780 781 vp = bp->b_vp; 782 if (vp != NULL) { 783 KASSERT(bp->b_objlock == vp->v_interlock); 784 if (vp->v_type == VBLK) 785 mp = vp->v_specmountpoint; 786 else 787 mp = vp->v_mount; 788 } else { 789 mp = NULL; 790 } 791 792 if (mp && mp->mnt_wapbl) { 793 if (bp->b_iodone != mp->mnt_wapbl_op->wo_wapbl_biodone) { 794 bdwrite(bp); 795 return 0; 796 } 797 } 798 799 /* 800 * Remember buffer type, to switch on it later. If the write was 801 * synchronous, but the file system was mounted with MNT_ASYNC, 802 * convert it to a delayed write. 803 * XXX note that this relies on delayed tape writes being converted 804 * to async, not sync writes (which is safe, but ugly). 805 */ 806 sync = !ISSET(bp->b_flags, B_ASYNC); 807 if (sync && mp != NULL && ISSET(mp->mnt_flag, MNT_ASYNC)) { 808 bdwrite(bp); 809 return (0); 810 } 811 812 /* 813 * Collect statistics on synchronous and asynchronous writes. 814 * Writes to block devices are charged to their associated 815 * filesystem (if any). 816 */ 817 if (mp != NULL) { 818 if (sync) 819 mp->mnt_stat.f_syncwrites++; 820 else 821 mp->mnt_stat.f_asyncwrites++; 822 } 823 824 /* 825 * Pay for the I/O operation and make sure the buf is on the correct 826 * vnode queue. 827 */ 828 bp->b_error = 0; 829 wasdelayed = ISSET(bp->b_oflags, BO_DELWRI); 830 CLR(bp->b_flags, B_READ); 831 if (wasdelayed) { 832 mutex_enter(&bufcache_lock); 833 mutex_enter(bp->b_objlock); 834 CLR(bp->b_oflags, BO_DONE | BO_DELWRI); 835 reassignbuf(bp, bp->b_vp); 836 mutex_exit(&bufcache_lock); 837 } else { 838 curlwp->l_ru.ru_oublock++; 839 mutex_enter(bp->b_objlock); 840 CLR(bp->b_oflags, BO_DONE | BO_DELWRI); 841 } 842 if (vp != NULL) 843 vp->v_numoutput++; 844 mutex_exit(bp->b_objlock); 845 846 /* Initiate disk write. */ 847 if (sync) 848 BIO_SETPRIO(bp, BPRIO_TIMECRITICAL); 849 else 850 BIO_SETPRIO(bp, BPRIO_TIMELIMITED); 851 852 VOP_STRATEGY(vp, bp); 853 854 if (sync) { 855 /* If I/O was synchronous, wait for it to complete. */ 856 rv = biowait(bp); 857 858 /* Release the buffer. */ 859 brelse(bp, 0); 860 861 return (rv); 862 } else { 863 return (0); 864 } 865 } 866 867 int 868 vn_bwrite(void *v) 869 { 870 struct vop_bwrite_args *ap = v; 871 872 return (bwrite(ap->a_bp)); 873 } 874 875 /* 876 * Delayed write. 877 * 878 * The buffer is marked dirty, but is not queued for I/O. 879 * This routine should be used when the buffer is expected 880 * to be modified again soon, typically a small write that 881 * partially fills a buffer. 882 * 883 * NB: magnetic tapes cannot be delayed; they must be 884 * written in the order that the writes are requested. 885 * 886 * Described in Leffler, et al. (pp. 208-213). 887 */ 888 void 889 bdwrite(buf_t *bp) 890 { 891 892 KASSERT(bp->b_vp == NULL || bp->b_vp->v_tag != VT_UFS || 893 bp->b_vp->v_type == VBLK || ISSET(bp->b_flags, B_COWDONE)); 894 KASSERT(ISSET(bp->b_cflags, BC_BUSY)); 895 KASSERT(!cv_has_waiters(&bp->b_done)); 896 897 /* If this is a tape block, write the block now. */ 898 if (bdev_type(bp->b_dev) == D_TAPE) { 899 bawrite(bp); 900 return; 901 } 902 903 if (wapbl_vphaswapbl(bp->b_vp)) { 904 struct mount *mp = wapbl_vptomp(bp->b_vp); 905 906 if (bp->b_iodone != mp->mnt_wapbl_op->wo_wapbl_biodone) { 907 WAPBL_ADD_BUF(mp, bp); 908 } 909 } 910 911 /* 912 * If the block hasn't been seen before: 913 * (1) Mark it as having been seen, 914 * (2) Charge for the write, 915 * (3) Make sure it's on its vnode's correct block list. 916 */ 917 KASSERT(bp->b_vp == NULL || bp->b_objlock == bp->b_vp->v_interlock); 918 919 if (!ISSET(bp->b_oflags, BO_DELWRI)) { 920 mutex_enter(&bufcache_lock); 921 mutex_enter(bp->b_objlock); 922 SET(bp->b_oflags, BO_DELWRI); 923 curlwp->l_ru.ru_oublock++; 924 reassignbuf(bp, bp->b_vp); 925 mutex_exit(&bufcache_lock); 926 } else { 927 mutex_enter(bp->b_objlock); 928 } 929 /* Otherwise, the "write" is done, so mark and release the buffer. */ 930 CLR(bp->b_oflags, BO_DONE); 931 mutex_exit(bp->b_objlock); 932 933 brelse(bp, 0); 934 } 935 936 /* 937 * Asynchronous block write; just an asynchronous bwrite(). 938 */ 939 void 940 bawrite(buf_t *bp) 941 { 942 943 KASSERT(ISSET(bp->b_cflags, BC_BUSY)); 944 KASSERT(bp->b_vp != NULL); 945 946 SET(bp->b_flags, B_ASYNC); 947 VOP_BWRITE(bp->b_vp, bp); 948 } 949 950 /* 951 * Release a buffer on to the free lists. 952 * Described in Bach (p. 46). 953 */ 954 void 955 brelsel(buf_t *bp, int set) 956 { 957 struct bqueue *bufq; 958 struct vnode *vp; 959 960 KASSERT(mutex_owned(&bufcache_lock)); 961 KASSERT(!cv_has_waiters(&bp->b_done)); 962 KASSERT(bp->b_refcnt > 0); 963 964 SET(bp->b_cflags, set); 965 966 KASSERT(ISSET(bp->b_cflags, BC_BUSY)); 967 KASSERT(bp->b_iodone == NULL); 968 969 /* Wake up any processes waiting for any buffer to become free. */ 970 cv_signal(&needbuffer_cv); 971 972 /* Wake up any proceeses waiting for _this_ buffer to become */ 973 if (ISSET(bp->b_cflags, BC_WANTED)) 974 CLR(bp->b_cflags, BC_WANTED|BC_AGE); 975 976 /* If it's clean clear the copy-on-write flag. */ 977 if (ISSET(bp->b_flags, B_COWDONE)) { 978 mutex_enter(bp->b_objlock); 979 if (!ISSET(bp->b_oflags, BO_DELWRI)) 980 CLR(bp->b_flags, B_COWDONE); 981 mutex_exit(bp->b_objlock); 982 } 983 984 /* 985 * Determine which queue the buffer should be on, then put it there. 986 */ 987 988 /* If it's locked, don't report an error; try again later. */ 989 if (ISSET(bp->b_flags, B_LOCKED)) 990 bp->b_error = 0; 991 992 /* If it's not cacheable, or an error, mark it invalid. */ 993 if (ISSET(bp->b_cflags, BC_NOCACHE) || bp->b_error != 0) 994 SET(bp->b_cflags, BC_INVAL); 995 996 if (ISSET(bp->b_cflags, BC_VFLUSH)) { 997 /* 998 * This is a delayed write buffer that was just flushed to 999 * disk. It is still on the LRU queue. If it's become 1000 * invalid, then we need to move it to a different queue; 1001 * otherwise leave it in its current position. 1002 */ 1003 CLR(bp->b_cflags, BC_VFLUSH); 1004 if (!ISSET(bp->b_cflags, BC_INVAL|BC_AGE) && 1005 !ISSET(bp->b_flags, B_LOCKED) && bp->b_error == 0) { 1006 KDASSERT(checkfreelist(bp, &bufqueues[BQ_LRU], 1)); 1007 goto already_queued; 1008 } else { 1009 bremfree(bp); 1010 } 1011 } 1012 1013 KDASSERT(checkfreelist(bp, &bufqueues[BQ_AGE], 0)); 1014 KDASSERT(checkfreelist(bp, &bufqueues[BQ_LRU], 0)); 1015 KDASSERT(checkfreelist(bp, &bufqueues[BQ_LOCKED], 0)); 1016 1017 if ((bp->b_bufsize <= 0) || ISSET(bp->b_cflags, BC_INVAL)) { 1018 /* 1019 * If it's invalid or empty, dissociate it from its vnode 1020 * and put on the head of the appropriate queue. 1021 */ 1022 if (ISSET(bp->b_flags, B_LOCKED)) { 1023 if (wapbl_vphaswapbl(vp = bp->b_vp)) { 1024 struct mount *mp = wapbl_vptomp(vp); 1025 1026 KASSERT(bp->b_iodone 1027 != mp->mnt_wapbl_op->wo_wapbl_biodone); 1028 WAPBL_REMOVE_BUF(mp, bp); 1029 } 1030 } 1031 1032 mutex_enter(bp->b_objlock); 1033 CLR(bp->b_oflags, BO_DONE|BO_DELWRI); 1034 if ((vp = bp->b_vp) != NULL) { 1035 KASSERT(bp->b_objlock == vp->v_interlock); 1036 reassignbuf(bp, bp->b_vp); 1037 brelvp(bp); 1038 mutex_exit(vp->v_interlock); 1039 } else { 1040 KASSERT(bp->b_objlock == &buffer_lock); 1041 mutex_exit(bp->b_objlock); 1042 } 1043 1044 if (bp->b_bufsize <= 0) 1045 /* no data */ 1046 goto already_queued; 1047 else 1048 /* invalid data */ 1049 bufq = &bufqueues[BQ_AGE]; 1050 binsheadfree(bp, bufq); 1051 } else { 1052 /* 1053 * It has valid data. Put it on the end of the appropriate 1054 * queue, so that it'll stick around for as long as possible. 1055 * If buf is AGE, but has dependencies, must put it on last 1056 * bufqueue to be scanned, ie LRU. This protects against the 1057 * livelock where BQ_AGE only has buffers with dependencies, 1058 * and we thus never get to the dependent buffers in BQ_LRU. 1059 */ 1060 if (ISSET(bp->b_flags, B_LOCKED)) { 1061 /* locked in core */ 1062 bufq = &bufqueues[BQ_LOCKED]; 1063 } else if (!ISSET(bp->b_cflags, BC_AGE)) { 1064 /* valid data */ 1065 bufq = &bufqueues[BQ_LRU]; 1066 } else { 1067 /* stale but valid data */ 1068 bufq = &bufqueues[BQ_AGE]; 1069 } 1070 binstailfree(bp, bufq); 1071 } 1072 already_queued: 1073 /* Unlock the buffer. */ 1074 CLR(bp->b_cflags, BC_AGE|BC_BUSY|BC_NOCACHE); 1075 CLR(bp->b_flags, B_ASYNC); 1076 cv_broadcast(&bp->b_busy); 1077 1078 if (bp->b_bufsize <= 0) 1079 brele(bp); 1080 } 1081 1082 void 1083 brelse(buf_t *bp, int set) 1084 { 1085 1086 mutex_enter(&bufcache_lock); 1087 brelsel(bp, set); 1088 mutex_exit(&bufcache_lock); 1089 } 1090 1091 /* 1092 * Determine if a block is in the cache. 1093 * Just look on what would be its hash chain. If it's there, return 1094 * a pointer to it, unless it's marked invalid. If it's marked invalid, 1095 * we normally don't return the buffer, unless the caller explicitly 1096 * wants us to. 1097 */ 1098 buf_t * 1099 incore(struct vnode *vp, daddr_t blkno) 1100 { 1101 buf_t *bp; 1102 1103 KASSERT(mutex_owned(&bufcache_lock)); 1104 1105 /* Search hash chain */ 1106 LIST_FOREACH(bp, BUFHASH(vp, blkno), b_hash) { 1107 if (bp->b_lblkno == blkno && bp->b_vp == vp && 1108 !ISSET(bp->b_cflags, BC_INVAL)) { 1109 KASSERT(bp->b_objlock == vp->v_interlock); 1110 return (bp); 1111 } 1112 } 1113 1114 return (NULL); 1115 } 1116 1117 /* 1118 * Get a block of requested size that is associated with 1119 * a given vnode and block offset. If it is found in the 1120 * block cache, mark it as having been found, make it busy 1121 * and return it. Otherwise, return an empty block of the 1122 * correct size. It is up to the caller to insure that the 1123 * cached blocks be of the correct size. 1124 */ 1125 buf_t * 1126 getblk(struct vnode *vp, daddr_t blkno, int size, int slpflag, int slptimeo) 1127 { 1128 int err, preserve; 1129 buf_t *bp; 1130 1131 mutex_enter(&bufcache_lock); 1132 loop: 1133 bp = incore(vp, blkno); 1134 if (bp != NULL) { 1135 err = bbusy(bp, ((slpflag & PCATCH) != 0), slptimeo, NULL); 1136 if (err != 0) { 1137 if (err == EPASSTHROUGH) 1138 goto loop; 1139 mutex_exit(&bufcache_lock); 1140 return (NULL); 1141 } 1142 KASSERT(!cv_has_waiters(&bp->b_done)); 1143 #ifdef DIAGNOSTIC 1144 if (ISSET(bp->b_oflags, BO_DONE|BO_DELWRI) && 1145 bp->b_bcount < size && vp->v_type != VBLK) 1146 panic("getblk: block size invariant failed"); 1147 #endif 1148 bremfree(bp); 1149 preserve = 1; 1150 } else { 1151 if ((bp = getnewbuf(slpflag, slptimeo, 0)) == NULL) 1152 goto loop; 1153 1154 if (incore(vp, blkno) != NULL) { 1155 /* The block has come into memory in the meantime. */ 1156 brelsel(bp, 0); 1157 goto loop; 1158 } 1159 1160 LIST_INSERT_HEAD(BUFHASH(vp, blkno), bp, b_hash); 1161 bp->b_blkno = bp->b_lblkno = bp->b_rawblkno = blkno; 1162 mutex_enter(vp->v_interlock); 1163 bgetvp(vp, bp); 1164 mutex_exit(vp->v_interlock); 1165 preserve = 0; 1166 } 1167 mutex_exit(&bufcache_lock); 1168 1169 /* 1170 * LFS can't track total size of B_LOCKED buffer (locked_queue_bytes) 1171 * if we re-size buffers here. 1172 */ 1173 if (ISSET(bp->b_flags, B_LOCKED)) { 1174 KASSERT(bp->b_bufsize >= size); 1175 } else { 1176 if (allocbuf(bp, size, preserve)) { 1177 mutex_enter(&bufcache_lock); 1178 LIST_REMOVE(bp, b_hash); 1179 mutex_exit(&bufcache_lock); 1180 brelse(bp, BC_INVAL); 1181 return NULL; 1182 } 1183 } 1184 BIO_SETPRIO(bp, BPRIO_DEFAULT); 1185 return (bp); 1186 } 1187 1188 /* 1189 * Get an empty, disassociated buffer of given size. 1190 */ 1191 buf_t * 1192 geteblk(int size) 1193 { 1194 buf_t *bp; 1195 int error; 1196 1197 mutex_enter(&bufcache_lock); 1198 while ((bp = getnewbuf(0, 0, 0)) == NULL) 1199 ; 1200 1201 SET(bp->b_cflags, BC_INVAL); 1202 LIST_INSERT_HEAD(&invalhash, bp, b_hash); 1203 mutex_exit(&bufcache_lock); 1204 BIO_SETPRIO(bp, BPRIO_DEFAULT); 1205 error = allocbuf(bp, size, 0); 1206 KASSERT(error == 0); 1207 return (bp); 1208 } 1209 1210 /* 1211 * Expand or contract the actual memory allocated to a buffer. 1212 * 1213 * If the buffer shrinks, data is lost, so it's up to the 1214 * caller to have written it out *first*; this routine will not 1215 * start a write. If the buffer grows, it's the callers 1216 * responsibility to fill out the buffer's additional contents. 1217 */ 1218 int 1219 allocbuf(buf_t *bp, int size, int preserve) 1220 { 1221 void *addr; 1222 vsize_t oldsize, desired_size; 1223 int oldcount; 1224 int delta; 1225 1226 desired_size = buf_roundsize(size); 1227 if (desired_size > MAXBSIZE) 1228 printf("allocbuf: buffer larger than MAXBSIZE requested"); 1229 1230 oldcount = bp->b_bcount; 1231 1232 bp->b_bcount = size; 1233 1234 oldsize = bp->b_bufsize; 1235 if (oldsize == desired_size) { 1236 /* 1237 * Do not short cut the WAPBL resize, as the buffer length 1238 * could still have changed and this would corrupt the 1239 * tracking of the transaction length. 1240 */ 1241 goto out; 1242 } 1243 1244 /* 1245 * If we want a buffer of a different size, re-allocate the 1246 * buffer's memory; copy old content only if needed. 1247 */ 1248 addr = buf_alloc(desired_size); 1249 if (addr == NULL) 1250 return ENOMEM; 1251 if (preserve) 1252 memcpy(addr, bp->b_data, MIN(oldsize,desired_size)); 1253 if (bp->b_data != NULL) 1254 buf_mrelease(bp->b_data, oldsize); 1255 bp->b_data = addr; 1256 bp->b_bufsize = desired_size; 1257 1258 /* 1259 * Update overall buffer memory counter (protected by bufcache_lock) 1260 */ 1261 delta = (long)desired_size - (long)oldsize; 1262 1263 mutex_enter(&bufcache_lock); 1264 if ((bufmem += delta) > bufmem_hiwater) { 1265 /* 1266 * Need to trim overall memory usage. 1267 */ 1268 while (buf_canrelease()) { 1269 if (curcpu()->ci_schedstate.spc_flags & 1270 SPCF_SHOULDYIELD) { 1271 mutex_exit(&bufcache_lock); 1272 preempt(); 1273 mutex_enter(&bufcache_lock); 1274 } 1275 if (buf_trim() == 0) 1276 break; 1277 } 1278 } 1279 mutex_exit(&bufcache_lock); 1280 1281 out: 1282 if (wapbl_vphaswapbl(bp->b_vp)) 1283 WAPBL_RESIZE_BUF(wapbl_vptomp(bp->b_vp), bp, oldsize, oldcount); 1284 1285 return 0; 1286 } 1287 1288 /* 1289 * Find a buffer which is available for use. 1290 * Select something from a free list. 1291 * Preference is to AGE list, then LRU list. 1292 * 1293 * Called with the buffer queues locked. 1294 * Return buffer locked. 1295 */ 1296 buf_t * 1297 getnewbuf(int slpflag, int slptimeo, int from_bufq) 1298 { 1299 buf_t *bp; 1300 struct vnode *vp; 1301 1302 start: 1303 KASSERT(mutex_owned(&bufcache_lock)); 1304 1305 /* 1306 * Get a new buffer from the pool. 1307 */ 1308 if (!from_bufq && buf_lotsfree()) { 1309 mutex_exit(&bufcache_lock); 1310 bp = pool_cache_get(buf_cache, PR_NOWAIT); 1311 if (bp != NULL) { 1312 memset((char *)bp, 0, sizeof(*bp)); 1313 buf_init(bp); 1314 SET(bp->b_cflags, BC_BUSY); /* mark buffer busy */ 1315 mutex_enter(&bufcache_lock); 1316 #if defined(DIAGNOSTIC) 1317 bp->b_freelistindex = -1; 1318 #endif /* defined(DIAGNOSTIC) */ 1319 return (bp); 1320 } 1321 mutex_enter(&bufcache_lock); 1322 } 1323 1324 KASSERT(mutex_owned(&bufcache_lock)); 1325 if ((bp = TAILQ_FIRST(&bufqueues[BQ_AGE].bq_queue)) != NULL || 1326 (bp = TAILQ_FIRST(&bufqueues[BQ_LRU].bq_queue)) != NULL) { 1327 KASSERT(!ISSET(bp->b_cflags, BC_BUSY) || ISSET(bp->b_cflags, BC_VFLUSH)); 1328 bremfree(bp); 1329 1330 /* Buffer is no longer on free lists. */ 1331 SET(bp->b_cflags, BC_BUSY); 1332 } else { 1333 /* 1334 * XXX: !from_bufq should be removed. 1335 */ 1336 if (!from_bufq || curlwp != uvm.pagedaemon_lwp) { 1337 /* wait for a free buffer of any kind */ 1338 if ((slpflag & PCATCH) != 0) 1339 (void)cv_timedwait_sig(&needbuffer_cv, 1340 &bufcache_lock, slptimeo); 1341 else 1342 (void)cv_timedwait(&needbuffer_cv, 1343 &bufcache_lock, slptimeo); 1344 } 1345 return (NULL); 1346 } 1347 1348 #ifdef DIAGNOSTIC 1349 if (bp->b_bufsize <= 0) 1350 panic("buffer %p: on queue but empty", bp); 1351 #endif 1352 1353 if (ISSET(bp->b_cflags, BC_VFLUSH)) { 1354 /* 1355 * This is a delayed write buffer being flushed to disk. Make 1356 * sure it gets aged out of the queue when it's finished, and 1357 * leave it off the LRU queue. 1358 */ 1359 CLR(bp->b_cflags, BC_VFLUSH); 1360 SET(bp->b_cflags, BC_AGE); 1361 goto start; 1362 } 1363 1364 KASSERT(ISSET(bp->b_cflags, BC_BUSY)); 1365 KASSERT(bp->b_refcnt > 0); 1366 KASSERT(!cv_has_waiters(&bp->b_done)); 1367 1368 /* 1369 * If buffer was a delayed write, start it and return NULL 1370 * (since we might sleep while starting the write). 1371 */ 1372 if (ISSET(bp->b_oflags, BO_DELWRI)) { 1373 /* 1374 * This buffer has gone through the LRU, so make sure it gets 1375 * reused ASAP. 1376 */ 1377 SET(bp->b_cflags, BC_AGE); 1378 mutex_exit(&bufcache_lock); 1379 bawrite(bp); 1380 mutex_enter(&bufcache_lock); 1381 return (NULL); 1382 } 1383 1384 vp = bp->b_vp; 1385 1386 /* clear out various other fields */ 1387 bp->b_cflags = BC_BUSY; 1388 bp->b_oflags = 0; 1389 bp->b_flags = 0; 1390 bp->b_dev = NODEV; 1391 bp->b_blkno = 0; 1392 bp->b_lblkno = 0; 1393 bp->b_rawblkno = 0; 1394 bp->b_iodone = 0; 1395 bp->b_error = 0; 1396 bp->b_resid = 0; 1397 bp->b_bcount = 0; 1398 1399 LIST_REMOVE(bp, b_hash); 1400 1401 /* Disassociate us from our vnode, if we had one... */ 1402 if (vp != NULL) { 1403 mutex_enter(vp->v_interlock); 1404 brelvp(bp); 1405 mutex_exit(vp->v_interlock); 1406 } 1407 1408 return (bp); 1409 } 1410 1411 /* 1412 * Attempt to free an aged buffer off the queues. 1413 * Called with queue lock held. 1414 * Returns the amount of buffer memory freed. 1415 */ 1416 static int 1417 buf_trim(void) 1418 { 1419 buf_t *bp; 1420 long size = 0; 1421 1422 KASSERT(mutex_owned(&bufcache_lock)); 1423 1424 /* Instruct getnewbuf() to get buffers off the queues */ 1425 if ((bp = getnewbuf(PCATCH, 1, 1)) == NULL) 1426 return 0; 1427 1428 KASSERT((bp->b_cflags & BC_WANTED) == 0); 1429 size = bp->b_bufsize; 1430 bufmem -= size; 1431 if (size > 0) { 1432 buf_mrelease(bp->b_data, size); 1433 bp->b_bcount = bp->b_bufsize = 0; 1434 } 1435 /* brelse() will return the buffer to the global buffer pool */ 1436 brelsel(bp, 0); 1437 return size; 1438 } 1439 1440 int 1441 buf_drain(int n) 1442 { 1443 int size = 0, sz; 1444 1445 KASSERT(mutex_owned(&bufcache_lock)); 1446 1447 while (size < n && bufmem > bufmem_lowater) { 1448 sz = buf_trim(); 1449 if (sz <= 0) 1450 break; 1451 size += sz; 1452 } 1453 1454 return size; 1455 } 1456 1457 /* 1458 * Wait for operations on the buffer to complete. 1459 * When they do, extract and return the I/O's error value. 1460 */ 1461 int 1462 biowait(buf_t *bp) 1463 { 1464 1465 KASSERT(ISSET(bp->b_cflags, BC_BUSY)); 1466 KASSERT(bp->b_refcnt > 0); 1467 1468 mutex_enter(bp->b_objlock); 1469 while (!ISSET(bp->b_oflags, BO_DONE | BO_DELWRI)) 1470 cv_wait(&bp->b_done, bp->b_objlock); 1471 mutex_exit(bp->b_objlock); 1472 1473 return bp->b_error; 1474 } 1475 1476 /* 1477 * Mark I/O complete on a buffer. 1478 * 1479 * If a callback has been requested, e.g. the pageout 1480 * daemon, do so. Otherwise, awaken waiting processes. 1481 * 1482 * [ Leffler, et al., says on p.247: 1483 * "This routine wakes up the blocked process, frees the buffer 1484 * for an asynchronous write, or, for a request by the pagedaemon 1485 * process, invokes a procedure specified in the buffer structure" ] 1486 * 1487 * In real life, the pagedaemon (or other system processes) wants 1488 * to do async stuff to, and doesn't want the buffer brelse()'d. 1489 * (for swap pager, that puts swap buffers on the free lists (!!!), 1490 * for the vn device, that puts allocated buffers on the free lists!) 1491 */ 1492 void 1493 biodone(buf_t *bp) 1494 { 1495 int s; 1496 1497 KASSERT(!ISSET(bp->b_oflags, BO_DONE)); 1498 1499 if (cpu_intr_p()) { 1500 /* From interrupt mode: defer to a soft interrupt. */ 1501 s = splvm(); 1502 TAILQ_INSERT_TAIL(&curcpu()->ci_data.cpu_biodone, bp, b_actq); 1503 softint_schedule(biodone_sih); 1504 splx(s); 1505 } else { 1506 /* Process now - the buffer may be freed soon. */ 1507 biodone2(bp); 1508 } 1509 } 1510 1511 static void 1512 biodone2(buf_t *bp) 1513 { 1514 void (*callout)(buf_t *); 1515 1516 mutex_enter(bp->b_objlock); 1517 /* Note that the transfer is done. */ 1518 if (ISSET(bp->b_oflags, BO_DONE)) 1519 panic("biodone2 already"); 1520 CLR(bp->b_flags, B_COWDONE); 1521 SET(bp->b_oflags, BO_DONE); 1522 BIO_SETPRIO(bp, BPRIO_DEFAULT); 1523 1524 /* Wake up waiting writers. */ 1525 if (!ISSET(bp->b_flags, B_READ)) 1526 vwakeup(bp); 1527 1528 if ((callout = bp->b_iodone) != NULL) { 1529 /* Note callout done, then call out. */ 1530 KASSERT(!cv_has_waiters(&bp->b_done)); 1531 KERNEL_LOCK(1, NULL); /* XXXSMP */ 1532 bp->b_iodone = NULL; 1533 mutex_exit(bp->b_objlock); 1534 (*callout)(bp); 1535 KERNEL_UNLOCK_ONE(NULL); /* XXXSMP */ 1536 } else if (ISSET(bp->b_flags, B_ASYNC)) { 1537 /* If async, release. */ 1538 KASSERT(!cv_has_waiters(&bp->b_done)); 1539 mutex_exit(bp->b_objlock); 1540 brelse(bp, 0); 1541 } else { 1542 /* Otherwise just wake up waiters in biowait(). */ 1543 cv_broadcast(&bp->b_done); 1544 mutex_exit(bp->b_objlock); 1545 } 1546 } 1547 1548 static void 1549 biointr(void *cookie) 1550 { 1551 struct cpu_info *ci; 1552 buf_t *bp; 1553 int s; 1554 1555 ci = curcpu(); 1556 1557 while (!TAILQ_EMPTY(&ci->ci_data.cpu_biodone)) { 1558 KASSERT(curcpu() == ci); 1559 1560 s = splvm(); 1561 bp = TAILQ_FIRST(&ci->ci_data.cpu_biodone); 1562 TAILQ_REMOVE(&ci->ci_data.cpu_biodone, bp, b_actq); 1563 splx(s); 1564 1565 biodone2(bp); 1566 } 1567 } 1568 1569 /* 1570 * Wait for all buffers to complete I/O 1571 * Return the number of "stuck" buffers. 1572 */ 1573 int 1574 buf_syncwait(void) 1575 { 1576 buf_t *bp; 1577 int iter, nbusy, nbusy_prev = 0, dcount, ihash; 1578 1579 dcount = 10000; 1580 for (iter = 0; iter < 20;) { 1581 mutex_enter(&bufcache_lock); 1582 nbusy = 0; 1583 for (ihash = 0; ihash < bufhash+1; ihash++) { 1584 LIST_FOREACH(bp, &bufhashtbl[ihash], b_hash) { 1585 if ((bp->b_cflags & (BC_BUSY|BC_INVAL)) == BC_BUSY) 1586 nbusy += ((bp->b_flags & B_READ) == 0); 1587 } 1588 } 1589 mutex_exit(&bufcache_lock); 1590 1591 if (nbusy == 0) 1592 break; 1593 if (nbusy_prev == 0) 1594 nbusy_prev = nbusy; 1595 printf("%d ", nbusy); 1596 kpause("bflush", false, MAX(1, hz / 25 * iter), NULL); 1597 if (nbusy >= nbusy_prev) /* we didn't flush anything */ 1598 iter++; 1599 else 1600 nbusy_prev = nbusy; 1601 } 1602 1603 if (nbusy) { 1604 #if defined(DEBUG) || defined(DEBUG_HALT_BUSY) 1605 printf("giving up\nPrinting vnodes for busy buffers\n"); 1606 for (ihash = 0; ihash < bufhash+1; ihash++) { 1607 LIST_FOREACH(bp, &bufhashtbl[ihash], b_hash) { 1608 if ((bp->b_cflags & (BC_BUSY|BC_INVAL)) == BC_BUSY && 1609 (bp->b_flags & B_READ) == 0) 1610 vprint(NULL, bp->b_vp); 1611 } 1612 } 1613 #endif 1614 } 1615 1616 return nbusy; 1617 } 1618 1619 static void 1620 sysctl_fillbuf(buf_t *i, struct buf_sysctl *o) 1621 { 1622 1623 o->b_flags = i->b_flags | i->b_cflags | i->b_oflags; 1624 o->b_error = i->b_error; 1625 o->b_prio = i->b_prio; 1626 o->b_dev = i->b_dev; 1627 o->b_bufsize = i->b_bufsize; 1628 o->b_bcount = i->b_bcount; 1629 o->b_resid = i->b_resid; 1630 o->b_addr = PTRTOUINT64(i->b_data); 1631 o->b_blkno = i->b_blkno; 1632 o->b_rawblkno = i->b_rawblkno; 1633 o->b_iodone = PTRTOUINT64(i->b_iodone); 1634 o->b_proc = PTRTOUINT64(i->b_proc); 1635 o->b_vp = PTRTOUINT64(i->b_vp); 1636 o->b_saveaddr = PTRTOUINT64(i->b_saveaddr); 1637 o->b_lblkno = i->b_lblkno; 1638 } 1639 1640 #define KERN_BUFSLOP 20 1641 static int 1642 sysctl_dobuf(SYSCTLFN_ARGS) 1643 { 1644 buf_t *bp; 1645 struct buf_sysctl bs; 1646 struct bqueue *bq; 1647 char *dp; 1648 u_int i, op, arg; 1649 size_t len, needed, elem_size, out_size; 1650 int error, elem_count, retries; 1651 1652 if (namelen == 1 && name[0] == CTL_QUERY) 1653 return (sysctl_query(SYSCTLFN_CALL(rnode))); 1654 1655 if (namelen != 4) 1656 return (EINVAL); 1657 1658 retries = 100; 1659 retry: 1660 dp = oldp; 1661 len = (oldp != NULL) ? *oldlenp : 0; 1662 op = name[0]; 1663 arg = name[1]; 1664 elem_size = name[2]; 1665 elem_count = name[3]; 1666 out_size = MIN(sizeof(bs), elem_size); 1667 1668 /* 1669 * at the moment, these are just "placeholders" to make the 1670 * API for retrieving kern.buf data more extensible in the 1671 * future. 1672 * 1673 * XXX kern.buf currently has "netbsd32" issues. hopefully 1674 * these will be resolved at a later point. 1675 */ 1676 if (op != KERN_BUF_ALL || arg != KERN_BUF_ALL || 1677 elem_size < 1 || elem_count < 0) 1678 return (EINVAL); 1679 1680 error = 0; 1681 needed = 0; 1682 sysctl_unlock(); 1683 mutex_enter(&bufcache_lock); 1684 for (i = 0; i < BQUEUES; i++) { 1685 bq = &bufqueues[i]; 1686 TAILQ_FOREACH(bp, &bq->bq_queue, b_freelist) { 1687 bq->bq_marker = bp; 1688 if (len >= elem_size && elem_count > 0) { 1689 sysctl_fillbuf(bp, &bs); 1690 mutex_exit(&bufcache_lock); 1691 error = copyout(&bs, dp, out_size); 1692 mutex_enter(&bufcache_lock); 1693 if (error) 1694 break; 1695 if (bq->bq_marker != bp) { 1696 /* 1697 * This sysctl node is only for 1698 * statistics. Retry; if the 1699 * queue keeps changing, then 1700 * bail out. 1701 */ 1702 if (retries-- == 0) { 1703 error = EAGAIN; 1704 break; 1705 } 1706 mutex_exit(&bufcache_lock); 1707 sysctl_relock(); 1708 goto retry; 1709 } 1710 dp += elem_size; 1711 len -= elem_size; 1712 } 1713 needed += elem_size; 1714 if (elem_count > 0 && elem_count != INT_MAX) 1715 elem_count--; 1716 } 1717 if (error != 0) 1718 break; 1719 } 1720 mutex_exit(&bufcache_lock); 1721 sysctl_relock(); 1722 1723 *oldlenp = needed; 1724 if (oldp == NULL) 1725 *oldlenp += KERN_BUFSLOP * sizeof(buf_t); 1726 1727 return (error); 1728 } 1729 1730 static int 1731 sysctl_bufvm_update(SYSCTLFN_ARGS) 1732 { 1733 int t, error, rv; 1734 struct sysctlnode node; 1735 1736 node = *rnode; 1737 node.sysctl_data = &t; 1738 t = *(int *)rnode->sysctl_data; 1739 error = sysctl_lookup(SYSCTLFN_CALL(&node)); 1740 if (error || newp == NULL) 1741 return (error); 1742 1743 if (t < 0) 1744 return EINVAL; 1745 if (rnode->sysctl_data == &bufcache) { 1746 if (t > 100) 1747 return (EINVAL); 1748 bufcache = t; 1749 buf_setwm(); 1750 } else if (rnode->sysctl_data == &bufmem_lowater) { 1751 if (bufmem_hiwater - t < 16) 1752 return (EINVAL); 1753 bufmem_lowater = t; 1754 } else if (rnode->sysctl_data == &bufmem_hiwater) { 1755 if (t - bufmem_lowater < 16) 1756 return (EINVAL); 1757 bufmem_hiwater = t; 1758 } else 1759 return (EINVAL); 1760 1761 /* Drain until below new high water mark */ 1762 sysctl_unlock(); 1763 mutex_enter(&bufcache_lock); 1764 while ((t = bufmem - bufmem_hiwater) >= 0) { 1765 rv = buf_drain(t / (2 * 1024)); 1766 if (rv <= 0) 1767 break; 1768 } 1769 mutex_exit(&bufcache_lock); 1770 sysctl_relock(); 1771 1772 return 0; 1773 } 1774 1775 static struct sysctllog *vfsbio_sysctllog; 1776 1777 static void 1778 sysctl_kern_buf_setup(void) 1779 { 1780 1781 sysctl_createv(&vfsbio_sysctllog, 0, NULL, NULL, 1782 CTLFLAG_PERMANENT, 1783 CTLTYPE_NODE, "kern", NULL, 1784 NULL, 0, NULL, 0, 1785 CTL_KERN, CTL_EOL); 1786 sysctl_createv(&vfsbio_sysctllog, 0, NULL, NULL, 1787 CTLFLAG_PERMANENT, 1788 CTLTYPE_NODE, "buf", 1789 SYSCTL_DESCR("Kernel buffer cache information"), 1790 sysctl_dobuf, 0, NULL, 0, 1791 CTL_KERN, KERN_BUF, CTL_EOL); 1792 } 1793 1794 static void 1795 sysctl_vm_buf_setup(void) 1796 { 1797 1798 sysctl_createv(&vfsbio_sysctllog, 0, NULL, NULL, 1799 CTLFLAG_PERMANENT, 1800 CTLTYPE_NODE, "vm", NULL, 1801 NULL, 0, NULL, 0, 1802 CTL_VM, CTL_EOL); 1803 sysctl_createv(&vfsbio_sysctllog, 0, NULL, NULL, 1804 CTLFLAG_PERMANENT|CTLFLAG_READWRITE, 1805 CTLTYPE_INT, "bufcache", 1806 SYSCTL_DESCR("Percentage of physical memory to use for " 1807 "buffer cache"), 1808 sysctl_bufvm_update, 0, &bufcache, 0, 1809 CTL_VM, CTL_CREATE, CTL_EOL); 1810 sysctl_createv(&vfsbio_sysctllog, 0, NULL, NULL, 1811 CTLFLAG_PERMANENT|CTLFLAG_READONLY, 1812 CTLTYPE_INT, "bufmem", 1813 SYSCTL_DESCR("Amount of kernel memory used by buffer " 1814 "cache"), 1815 NULL, 0, &bufmem, 0, 1816 CTL_VM, CTL_CREATE, CTL_EOL); 1817 sysctl_createv(&vfsbio_sysctllog, 0, NULL, NULL, 1818 CTLFLAG_PERMANENT|CTLFLAG_READWRITE, 1819 CTLTYPE_INT, "bufmem_lowater", 1820 SYSCTL_DESCR("Minimum amount of kernel memory to " 1821 "reserve for buffer cache"), 1822 sysctl_bufvm_update, 0, &bufmem_lowater, 0, 1823 CTL_VM, CTL_CREATE, CTL_EOL); 1824 sysctl_createv(&vfsbio_sysctllog, 0, NULL, NULL, 1825 CTLFLAG_PERMANENT|CTLFLAG_READWRITE, 1826 CTLTYPE_INT, "bufmem_hiwater", 1827 SYSCTL_DESCR("Maximum amount of kernel memory to use " 1828 "for buffer cache"), 1829 sysctl_bufvm_update, 0, &bufmem_hiwater, 0, 1830 CTL_VM, CTL_CREATE, CTL_EOL); 1831 } 1832 1833 #ifdef DEBUG 1834 /* 1835 * Print out statistics on the current allocation of the buffer pool. 1836 * Can be enabled to print out on every ``sync'' by setting "syncprt" 1837 * in vfs_syscalls.c using sysctl. 1838 */ 1839 void 1840 vfs_bufstats(void) 1841 { 1842 int i, j, count; 1843 buf_t *bp; 1844 struct bqueue *dp; 1845 int counts[(MAXBSIZE / PAGE_SIZE) + 1]; 1846 static const char *bname[BQUEUES] = { "LOCKED", "LRU", "AGE" }; 1847 1848 for (dp = bufqueues, i = 0; dp < &bufqueues[BQUEUES]; dp++, i++) { 1849 count = 0; 1850 for (j = 0; j <= MAXBSIZE/PAGE_SIZE; j++) 1851 counts[j] = 0; 1852 TAILQ_FOREACH(bp, &dp->bq_queue, b_freelist) { 1853 counts[bp->b_bufsize/PAGE_SIZE]++; 1854 count++; 1855 } 1856 printf("%s: total-%d", bname[i], count); 1857 for (j = 0; j <= MAXBSIZE/PAGE_SIZE; j++) 1858 if (counts[j] != 0) 1859 printf(", %d-%d", j * PAGE_SIZE, counts[j]); 1860 printf("\n"); 1861 } 1862 } 1863 #endif /* DEBUG */ 1864 1865 /* ------------------------------ */ 1866 1867 buf_t * 1868 getiobuf(struct vnode *vp, bool waitok) 1869 { 1870 buf_t *bp; 1871 1872 bp = pool_cache_get(bufio_cache, (waitok ? PR_WAITOK : PR_NOWAIT)); 1873 if (bp == NULL) 1874 return bp; 1875 1876 buf_init(bp); 1877 1878 if ((bp->b_vp = vp) == NULL) 1879 bp->b_objlock = &buffer_lock; 1880 else 1881 bp->b_objlock = vp->v_interlock; 1882 1883 return bp; 1884 } 1885 1886 void 1887 putiobuf(buf_t *bp) 1888 { 1889 1890 buf_destroy(bp); 1891 pool_cache_put(bufio_cache, bp); 1892 } 1893 1894 /* 1895 * nestiobuf_iodone: b_iodone callback for nested buffers. 1896 */ 1897 1898 void 1899 nestiobuf_iodone(buf_t *bp) 1900 { 1901 buf_t *mbp = bp->b_private; 1902 int error; 1903 int donebytes; 1904 1905 KASSERT(bp->b_bcount <= bp->b_bufsize); 1906 KASSERT(mbp != bp); 1907 1908 error = bp->b_error; 1909 if (bp->b_error == 0 && 1910 (bp->b_bcount < bp->b_bufsize || bp->b_resid > 0)) { 1911 /* 1912 * Not all got transfered, raise an error. We have no way to 1913 * propagate these conditions to mbp. 1914 */ 1915 error = EIO; 1916 } 1917 1918 donebytes = bp->b_bufsize; 1919 1920 putiobuf(bp); 1921 nestiobuf_done(mbp, donebytes, error); 1922 } 1923 1924 /* 1925 * nestiobuf_setup: setup a "nested" buffer. 1926 * 1927 * => 'mbp' is a "master" buffer which is being divided into sub pieces. 1928 * => 'bp' should be a buffer allocated by getiobuf. 1929 * => 'offset' is a byte offset in the master buffer. 1930 * => 'size' is a size in bytes of this nested buffer. 1931 */ 1932 1933 void 1934 nestiobuf_setup(buf_t *mbp, buf_t *bp, int offset, size_t size) 1935 { 1936 const int b_read = mbp->b_flags & B_READ; 1937 struct vnode *vp = mbp->b_vp; 1938 1939 KASSERT(mbp->b_bcount >= offset + size); 1940 bp->b_vp = vp; 1941 bp->b_dev = mbp->b_dev; 1942 bp->b_objlock = mbp->b_objlock; 1943 bp->b_cflags = BC_BUSY; 1944 bp->b_flags = B_ASYNC | b_read; 1945 bp->b_iodone = nestiobuf_iodone; 1946 bp->b_data = (char *)mbp->b_data + offset; 1947 bp->b_resid = bp->b_bcount = size; 1948 bp->b_bufsize = bp->b_bcount; 1949 bp->b_private = mbp; 1950 BIO_COPYPRIO(bp, mbp); 1951 if (!b_read && vp != NULL) { 1952 mutex_enter(vp->v_interlock); 1953 vp->v_numoutput++; 1954 mutex_exit(vp->v_interlock); 1955 } 1956 } 1957 1958 /* 1959 * nestiobuf_done: propagate completion to the master buffer. 1960 * 1961 * => 'donebytes' specifies how many bytes in the 'mbp' is completed. 1962 * => 'error' is an errno(2) that 'donebytes' has been completed with. 1963 */ 1964 1965 void 1966 nestiobuf_done(buf_t *mbp, int donebytes, int error) 1967 { 1968 1969 if (donebytes == 0) { 1970 return; 1971 } 1972 mutex_enter(mbp->b_objlock); 1973 KASSERT(mbp->b_resid >= donebytes); 1974 mbp->b_resid -= donebytes; 1975 if (error) 1976 mbp->b_error = error; 1977 if (mbp->b_resid == 0) { 1978 if (mbp->b_error) 1979 mbp->b_resid = mbp->b_bcount; 1980 mutex_exit(mbp->b_objlock); 1981 biodone(mbp); 1982 } else 1983 mutex_exit(mbp->b_objlock); 1984 } 1985 1986 void 1987 buf_init(buf_t *bp) 1988 { 1989 1990 cv_init(&bp->b_busy, "biolock"); 1991 cv_init(&bp->b_done, "biowait"); 1992 bp->b_dev = NODEV; 1993 bp->b_error = 0; 1994 bp->b_flags = 0; 1995 bp->b_cflags = 0; 1996 bp->b_oflags = 0; 1997 bp->b_objlock = &buffer_lock; 1998 bp->b_iodone = NULL; 1999 bp->b_refcnt = 1; 2000 bp->b_dev = NODEV; 2001 bp->b_vnbufs.le_next = NOLIST; 2002 BIO_SETPRIO(bp, BPRIO_DEFAULT); 2003 } 2004 2005 void 2006 buf_destroy(buf_t *bp) 2007 { 2008 2009 cv_destroy(&bp->b_done); 2010 cv_destroy(&bp->b_busy); 2011 } 2012 2013 int 2014 bbusy(buf_t *bp, bool intr, int timo, kmutex_t *interlock) 2015 { 2016 int error; 2017 2018 KASSERT(mutex_owned(&bufcache_lock)); 2019 2020 if ((bp->b_cflags & BC_BUSY) != 0) { 2021 if (curlwp == uvm.pagedaemon_lwp) 2022 return EDEADLK; 2023 bp->b_cflags |= BC_WANTED; 2024 bref(bp); 2025 if (interlock != NULL) 2026 mutex_exit(interlock); 2027 if (intr) { 2028 error = cv_timedwait_sig(&bp->b_busy, &bufcache_lock, 2029 timo); 2030 } else { 2031 error = cv_timedwait(&bp->b_busy, &bufcache_lock, 2032 timo); 2033 } 2034 brele(bp); 2035 if (interlock != NULL) 2036 mutex_enter(interlock); 2037 if (error != 0) 2038 return error; 2039 return EPASSTHROUGH; 2040 } 2041 bp->b_cflags |= BC_BUSY; 2042 2043 return 0; 2044 } 2045