1 /* $NetBSD: vfs_bio.c,v 1.306 2024/12/07 02:27:38 riastradh Exp $ */ 2 3 /*- 4 * Copyright (c) 2007, 2008, 2009, 2019, 2020 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 1.306 2024/12/07 02:27:38 riastradh Exp $"); 127 128 #ifdef _KERNEL_OPT 129 #include "opt_biohist.h" 130 #include "opt_bufcache.h" 131 #include "opt_dtrace.h" 132 #endif 133 134 #include <sys/param.h> 135 #include <sys/types.h> 136 137 #include <sys/bitops.h> 138 #include <sys/buf.h> 139 #include <sys/conf.h> 140 #include <sys/cprng.h> 141 #include <sys/cpu.h> 142 #include <sys/fstrans.h> 143 #include <sys/intr.h> 144 #include <sys/kauth.h> 145 #include <sys/kernel.h> 146 #include <sys/mount.h> 147 #include <sys/proc.h> 148 #include <sys/resourcevar.h> 149 #include <sys/sdt.h> 150 #include <sys/sysctl.h> 151 #include <sys/systm.h> 152 #include <sys/vnode.h> 153 #include <sys/wapbl.h> 154 155 #include <uvm/uvm.h> /* extern struct uvm uvm */ 156 157 #include <miscfs/specfs/specdev.h> 158 159 SDT_PROVIDER_DEFINE(io); 160 161 SDT_PROBE_DEFINE4(io, kernel, , bbusy__start, 162 "struct buf *"/*bp*/, 163 "bool"/*intr*/, "int"/*timo*/, "kmutex_t *"/*interlock*/); 164 SDT_PROBE_DEFINE5(io, kernel, , bbusy__done, 165 "struct buf *"/*bp*/, 166 "bool"/*intr*/, 167 "int"/*timo*/, 168 "kmutex_t *"/*interlock*/, 169 "int"/*error*/); 170 SDT_PROBE_DEFINE0(io, kernel, , getnewbuf__start); 171 SDT_PROBE_DEFINE1(io, kernel, , getnewbuf__done, "struct buf *"/*bp*/); 172 SDT_PROBE_DEFINE3(io, kernel, , getblk__start, 173 "struct vnode *"/*vp*/, "daddr_t"/*blkno*/, "int"/*size*/); 174 SDT_PROBE_DEFINE4(io, kernel, , getblk__done, 175 "struct vnode *"/*vp*/, "daddr_t"/*blkno*/, "int"/*size*/, 176 "struct buf *"/*bp*/); 177 SDT_PROBE_DEFINE2(io, kernel, , brelse, "struct buf *"/*bp*/, "int"/*set*/); 178 SDT_PROBE_DEFINE1(io, kernel, , wait__start, "struct buf *"/*bp*/); 179 SDT_PROBE_DEFINE1(io, kernel, , wait__done, "struct buf *"/*bp*/); 180 181 #ifndef BUFPAGES 182 # define BUFPAGES 0 183 #endif 184 185 #ifdef BUFCACHE 186 # if (BUFCACHE < 5) || (BUFCACHE > 95) 187 # error BUFCACHE is not between 5 and 95 188 # endif 189 #else 190 # define BUFCACHE 15 191 #endif 192 193 u_int nbuf; /* desired number of buffer headers */ 194 u_int bufpages = BUFPAGES; /* optional hardwired count */ 195 u_int bufcache = BUFCACHE; /* max % of RAM to use for buffer cache */ 196 197 /* 198 * Definitions for the buffer free lists. 199 */ 200 #define BQUEUES 3 /* number of free buffer queues */ 201 202 #define BQ_LOCKED 0 /* super-blocks &c */ 203 #define BQ_LRU 1 /* lru, useful buffers */ 204 #define BQ_AGE 2 /* rubbish */ 205 206 struct bqueue { 207 TAILQ_HEAD(, buf) bq_queue; 208 uint64_t bq_bytes; 209 buf_t *bq_marker; 210 }; 211 static struct bqueue bufqueues[BQUEUES] __cacheline_aligned; 212 213 /* Function prototypes */ 214 static void buf_setwm(void); 215 static int buf_trim(void); 216 static void *bufpool_page_alloc(struct pool *, int); 217 static void bufpool_page_free(struct pool *, void *); 218 static buf_t *bio_doread(struct vnode *, daddr_t, int, int); 219 static buf_t *getnewbuf(int, int, int); 220 static int buf_lotsfree(void); 221 static int buf_canrelease(void); 222 static u_long buf_mempoolidx(u_long); 223 static u_long buf_roundsize(u_long); 224 static void *buf_alloc(size_t); 225 static void buf_mrelease(void *, size_t); 226 static void binsheadfree(buf_t *, struct bqueue *); 227 static void binstailfree(buf_t *, struct bqueue *); 228 #ifdef DEBUG 229 static int checkfreelist(buf_t *, struct bqueue *, int); 230 #endif 231 static void biointr(void *); 232 static void biodone2(buf_t *); 233 static void sysctl_kern_buf_setup(void); 234 static void sysctl_vm_buf_setup(void); 235 236 /* Initialization for biohist */ 237 238 #include <sys/biohist.h> 239 240 BIOHIST_DEFINE(biohist); 241 242 void 243 biohist_init(void) 244 { 245 246 BIOHIST_INIT(biohist, BIOHIST_SIZE); 247 } 248 249 /* 250 * Definitions for the buffer hash lists. 251 */ 252 #define BUFHASH(dvp, lbn) \ 253 (&bufhashtbl[(((long)(dvp) >> 8) + (int)(lbn)) & bufhash]) 254 LIST_HEAD(bufhashhdr, buf) *bufhashtbl, invalhash; 255 u_long bufhash; 256 257 static int bufhash_stats(struct hashstat_sysctl *, bool); 258 259 static kcondvar_t needbuffer_cv; 260 261 /* 262 * Buffer queue lock. 263 */ 264 kmutex_t bufcache_lock __cacheline_aligned; 265 kmutex_t buffer_lock __cacheline_aligned; 266 267 /* Software ISR for completed transfers. */ 268 static void *biodone_sih; 269 270 /* Buffer pool for I/O buffers. */ 271 static pool_cache_t buf_cache; 272 static pool_cache_t bufio_cache; 273 274 #define MEMPOOL_INDEX_OFFSET (ilog2(DEV_BSIZE)) /* smallest pool is 512 bytes */ 275 #define NMEMPOOLS (ilog2(MAXBSIZE) - MEMPOOL_INDEX_OFFSET + 1) 276 __CTASSERT((1 << (NMEMPOOLS + MEMPOOL_INDEX_OFFSET - 1)) == MAXBSIZE); 277 278 /* Buffer memory pools */ 279 static struct pool bmempools[NMEMPOOLS]; 280 281 static struct vm_map *buf_map; 282 283 /* 284 * Buffer memory pool allocator. 285 */ 286 static void * 287 bufpool_page_alloc(struct pool *pp, int flags) 288 { 289 290 return (void *)uvm_km_alloc(buf_map, 291 MAXBSIZE, MAXBSIZE, 292 ((flags & PR_WAITOK) ? 0 : UVM_KMF_NOWAIT|UVM_KMF_TRYLOCK) 293 | UVM_KMF_WIRED); 294 } 295 296 static void 297 bufpool_page_free(struct pool *pp, void *v) 298 { 299 300 uvm_km_free(buf_map, (vaddr_t)v, MAXBSIZE, UVM_KMF_WIRED); 301 } 302 303 static struct pool_allocator bufmempool_allocator = { 304 .pa_alloc = bufpool_page_alloc, 305 .pa_free = bufpool_page_free, 306 .pa_pagesz = MAXBSIZE, 307 }; 308 309 /* Buffer memory management variables */ 310 u_long bufmem_valimit; 311 u_long bufmem_hiwater; 312 u_long bufmem_lowater; 313 u_long bufmem; 314 315 /* 316 * MD code can call this to set a hard limit on the amount 317 * of virtual memory used by the buffer cache. 318 */ 319 int 320 buf_setvalimit(vsize_t sz) 321 { 322 323 /* We need to accommodate at least NMEMPOOLS of MAXBSIZE each */ 324 if (sz < NMEMPOOLS * MAXBSIZE) 325 return SET_ERROR(EINVAL); 326 327 bufmem_valimit = sz; 328 return 0; 329 } 330 331 static void 332 buf_setwm(void) 333 { 334 335 bufmem_hiwater = buf_memcalc(); 336 /* lowater is approx. 2% of memory (with bufcache = 15) */ 337 #define BUFMEM_WMSHIFT 3 338 #define BUFMEM_HIWMMIN (64 * 1024 << BUFMEM_WMSHIFT) 339 if (bufmem_hiwater < BUFMEM_HIWMMIN) 340 /* Ensure a reasonable minimum value */ 341 bufmem_hiwater = BUFMEM_HIWMMIN; 342 bufmem_lowater = bufmem_hiwater >> BUFMEM_WMSHIFT; 343 } 344 345 #ifdef DEBUG 346 int debug_verify_freelist = 0; 347 static int 348 checkfreelist(buf_t *bp, struct bqueue *dp, int ison) 349 { 350 buf_t *b; 351 352 if (!debug_verify_freelist) 353 return 1; 354 355 TAILQ_FOREACH(b, &dp->bq_queue, b_freelist) { 356 if (b == bp) 357 return ison ? 1 : 0; 358 } 359 360 return ison ? 0 : 1; 361 } 362 #endif 363 364 /* 365 * Insq/Remq for the buffer hash lists. 366 * Call with buffer queue locked. 367 */ 368 static void 369 binsheadfree(buf_t *bp, struct bqueue *dp) 370 { 371 372 KASSERT(mutex_owned(&bufcache_lock)); 373 KASSERT(bp->b_freelistindex == -1); 374 TAILQ_INSERT_HEAD(&dp->bq_queue, bp, b_freelist); 375 dp->bq_bytes += bp->b_bufsize; 376 bp->b_freelistindex = dp - bufqueues; 377 } 378 379 static void 380 binstailfree(buf_t *bp, struct bqueue *dp) 381 { 382 383 KASSERT(mutex_owned(&bufcache_lock)); 384 KASSERTMSG(bp->b_freelistindex == -1, "double free of buffer? " 385 "bp=%p, b_freelistindex=%d\n", bp, bp->b_freelistindex); 386 TAILQ_INSERT_TAIL(&dp->bq_queue, bp, b_freelist); 387 dp->bq_bytes += bp->b_bufsize; 388 bp->b_freelistindex = dp - bufqueues; 389 } 390 391 void 392 bremfree(buf_t *bp) 393 { 394 struct bqueue *dp; 395 int bqidx = bp->b_freelistindex; 396 397 KASSERT(mutex_owned(&bufcache_lock)); 398 399 KASSERT(bqidx != -1); 400 dp = &bufqueues[bqidx]; 401 KDASSERT(checkfreelist(bp, dp, 1)); 402 KASSERT(dp->bq_bytes >= bp->b_bufsize); 403 TAILQ_REMOVE(&dp->bq_queue, bp, b_freelist); 404 dp->bq_bytes -= bp->b_bufsize; 405 406 /* For the sysctl helper. */ 407 if (bp == dp->bq_marker) 408 dp->bq_marker = NULL; 409 410 #if defined(DIAGNOSTIC) 411 bp->b_freelistindex = -1; 412 #endif /* defined(DIAGNOSTIC) */ 413 } 414 415 /* 416 * note that for some ports this is used by pmap bootstrap code to 417 * determine kva size. 418 */ 419 u_long 420 buf_memcalc(void) 421 { 422 u_long n; 423 vsize_t mapsz = 0; 424 425 /* 426 * Determine the upper bound of memory to use for buffers. 427 * 428 * - If bufpages is specified, use that as the number 429 * pages. 430 * 431 * - Otherwise, use bufcache as the percentage of 432 * physical memory. 433 */ 434 if (bufpages != 0) { 435 n = bufpages; 436 } else { 437 if (bufcache < 5) { 438 printf("forcing bufcache %d -> 5", bufcache); 439 bufcache = 5; 440 } 441 if (bufcache > 95) { 442 printf("forcing bufcache %d -> 95", bufcache); 443 bufcache = 95; 444 } 445 if (buf_map != NULL) 446 mapsz = vm_map_max(buf_map) - vm_map_min(buf_map); 447 n = calc_cache_size(mapsz, bufcache, 448 (buf_map != kernel_map) ? 100 : BUFCACHE_VA_MAXPCT) 449 / PAGE_SIZE; 450 } 451 452 n <<= PAGE_SHIFT; 453 if (bufmem_valimit != 0 && n > bufmem_valimit) 454 n = bufmem_valimit; 455 456 return n; 457 } 458 459 /* 460 * Initialize buffers and hash links for buffers. 461 */ 462 void 463 bufinit(void) 464 { 465 struct bqueue *dp; 466 int use_std; 467 u_int i; 468 469 biodone_vfs = biodone; 470 471 mutex_init(&bufcache_lock, MUTEX_DEFAULT, IPL_NONE); 472 mutex_init(&buffer_lock, MUTEX_DEFAULT, IPL_NONE); 473 cv_init(&needbuffer_cv, "needbuf"); 474 475 if (bufmem_valimit != 0) { 476 vaddr_t minaddr = 0, maxaddr; 477 buf_map = uvm_km_suballoc(kernel_map, &minaddr, &maxaddr, 478 bufmem_valimit, 0, false, 0); 479 if (buf_map == NULL) 480 panic("bufinit: cannot allocate submap"); 481 } else 482 buf_map = kernel_map; 483 484 /* 485 * Initialize buffer cache memory parameters. 486 */ 487 bufmem = 0; 488 buf_setwm(); 489 490 /* On "small" machines use small pool page sizes where possible */ 491 use_std = (physmem < atop(16*1024*1024)); 492 493 /* 494 * Also use them on systems that can map the pool pages using 495 * a direct-mapped segment. 496 */ 497 #ifdef PMAP_MAP_POOLPAGE 498 use_std = 1; 499 #endif 500 501 buf_cache = pool_cache_init(sizeof(buf_t), 0, 0, 0, 502 "bufpl", NULL, IPL_SOFTBIO, NULL, NULL, NULL); 503 bufio_cache = pool_cache_init(sizeof(buf_t), 0, 0, 0, 504 "biopl", NULL, IPL_BIO, NULL, NULL, NULL); 505 506 for (i = 0; i < NMEMPOOLS; i++) { 507 struct pool_allocator *pa; 508 struct pool *pp = &bmempools[i]; 509 u_int size = 1 << (i + MEMPOOL_INDEX_OFFSET); 510 char *name = kmem_alloc(8, KM_SLEEP); /* XXX: never freed */ 511 512 if (__predict_false(size >= 1048576)) 513 (void)snprintf(name, 8, "buf%um", size / 1048576); 514 else if (__predict_true(size >= 1024)) 515 (void)snprintf(name, 8, "buf%uk", size / 1024); 516 else 517 (void)snprintf(name, 8, "buf%ub", size); 518 pa = (size <= PAGE_SIZE && use_std) 519 ? &pool_allocator_nointr 520 : &bufmempool_allocator; 521 pool_init(pp, size, DEV_BSIZE, 0, 0, name, pa, IPL_NONE); 522 pool_setlowat(pp, 1); 523 pool_sethiwat(pp, 1); 524 } 525 526 /* Initialize the buffer queues */ 527 for (dp = bufqueues; dp < &bufqueues[BQUEUES]; dp++) { 528 TAILQ_INIT(&dp->bq_queue); 529 dp->bq_bytes = 0; 530 } 531 532 /* 533 * Estimate hash table size based on the amount of memory we 534 * intend to use for the buffer cache. The average buffer 535 * size is dependent on our clients (i.e. filesystems). 536 * 537 * For now, use an empirical 3K per buffer. 538 */ 539 nbuf = (bufmem_hiwater / 1024) / 3; 540 bufhashtbl = hashinit(nbuf, HASH_LIST, true, &bufhash); 541 542 sysctl_kern_buf_setup(); 543 sysctl_vm_buf_setup(); 544 hashstat_register("bufhash", bufhash_stats); 545 } 546 547 void 548 bufinit2(void) 549 { 550 551 biodone_sih = softint_establish(SOFTINT_BIO | SOFTINT_MPSAFE, biointr, 552 NULL); 553 if (biodone_sih == NULL) 554 panic("bufinit2: can't establish soft interrupt"); 555 } 556 557 static int 558 buf_lotsfree(void) 559 { 560 u_long guess; 561 562 /* Always allocate if less than the low water mark. */ 563 if (bufmem < bufmem_lowater) 564 return 1; 565 566 /* Never allocate if greater than the high water mark. */ 567 if (bufmem > bufmem_hiwater) 568 return 0; 569 570 /* If there's anything on the AGE list, it should be eaten. */ 571 if (TAILQ_FIRST(&bufqueues[BQ_AGE].bq_queue) != NULL) 572 return 0; 573 574 /* 575 * The probabily of getting a new allocation is inversely 576 * proportional to the current size of the cache above 577 * the low water mark. Divide the total first to avoid overflows 578 * in the product. 579 */ 580 guess = cprng_fast32() % 16; 581 582 if ((bufmem_hiwater - bufmem_lowater) / 16 * guess >= 583 (bufmem - bufmem_lowater)) 584 return 1; 585 586 /* Otherwise don't allocate. */ 587 return 0; 588 } 589 590 /* 591 * Return estimate of bytes we think need to be 592 * released to help resolve low memory conditions. 593 * 594 * => called with bufcache_lock held. 595 */ 596 static int 597 buf_canrelease(void) 598 { 599 int pagedemand, ninvalid = 0; 600 601 KASSERT(mutex_owned(&bufcache_lock)); 602 603 if (bufmem < bufmem_lowater) 604 return 0; 605 606 if (bufmem > bufmem_hiwater) 607 return bufmem - bufmem_hiwater; 608 609 ninvalid += bufqueues[BQ_AGE].bq_bytes; 610 611 pagedemand = uvmexp.freetarg - uvm_availmem(false); 612 if (pagedemand < 0) 613 return ninvalid; 614 return MAX(ninvalid, MIN(2 * MAXBSIZE, 615 MIN((bufmem - bufmem_lowater) / 16, pagedemand * PAGE_SIZE))); 616 } 617 618 /* 619 * Buffer memory allocation helper functions 620 */ 621 static u_long 622 buf_mempoolidx(u_long size) 623 { 624 u_int n = 0; 625 626 size -= 1; 627 size >>= MEMPOOL_INDEX_OFFSET; 628 while (size) { 629 size >>= 1; 630 n += 1; 631 } 632 if (n >= NMEMPOOLS) 633 panic("buf mem pool index %d", n); 634 return n; 635 } 636 637 static u_long 638 buf_roundsize(u_long size) 639 { 640 641 /* Round up to nearest power of 2 */ 642 return (1 << (buf_mempoolidx(size) + MEMPOOL_INDEX_OFFSET)); 643 } 644 645 static void * 646 buf_alloc(size_t size) 647 { 648 u_int n = buf_mempoolidx(size); 649 void *addr; 650 651 while (1) { 652 addr = pool_get(&bmempools[n], PR_NOWAIT); 653 if (addr != NULL) 654 break; 655 656 /* No memory, see if we can free some. If so, try again */ 657 mutex_enter(&bufcache_lock); 658 if (buf_drain(1) > 0) { 659 mutex_exit(&bufcache_lock); 660 continue; 661 } 662 663 if (curlwp == uvm.pagedaemon_lwp) { 664 mutex_exit(&bufcache_lock); 665 return NULL; 666 } 667 668 /* Wait for buffers to arrive on the LRU queue */ 669 cv_timedwait(&needbuffer_cv, &bufcache_lock, hz / 4); 670 mutex_exit(&bufcache_lock); 671 } 672 673 return addr; 674 } 675 676 static void 677 buf_mrelease(void *addr, size_t size) 678 { 679 680 pool_put(&bmempools[buf_mempoolidx(size)], addr); 681 } 682 683 /* 684 * bread()/breadn() helper. 685 */ 686 static buf_t * 687 bio_doread(struct vnode *vp, daddr_t blkno, int size, int async) 688 { 689 buf_t *bp; 690 struct mount *mp; 691 692 bp = getblk(vp, blkno, size, 0, 0); 693 694 /* 695 * getblk() may return NULL if we are the pagedaemon. 696 */ 697 if (bp == NULL) { 698 KASSERT(curlwp == uvm.pagedaemon_lwp); 699 return NULL; 700 } 701 702 /* 703 * If buffer does not have data valid, start a read. 704 * Note that if buffer is BC_INVAL, getblk() won't return it. 705 * Therefore, it's valid if its I/O has completed or been delayed. 706 */ 707 if (!ISSET(bp->b_oflags, (BO_DONE | BO_DELWRI))) { 708 /* Start I/O for the buffer. */ 709 SET(bp->b_flags, B_READ | async); 710 if (async) 711 BIO_SETPRIO(bp, BPRIO_TIMELIMITED); 712 else 713 BIO_SETPRIO(bp, BPRIO_TIMECRITICAL); 714 VOP_STRATEGY(vp, bp); 715 716 /* Pay for the read. */ 717 curlwp->l_ru.ru_inblock++; 718 } else if (async) 719 brelse(bp, 0); 720 721 if (vp->v_type == VBLK) 722 mp = spec_node_getmountedfs(vp); 723 else 724 mp = vp->v_mount; 725 726 /* 727 * Collect statistics on synchronous and asynchronous reads. 728 * Reads from block devices are charged to their associated 729 * filesystem (if any). 730 */ 731 if (mp != NULL) { 732 if (async == 0) 733 mp->mnt_stat.f_syncreads++; 734 else 735 mp->mnt_stat.f_asyncreads++; 736 } 737 738 return bp; 739 } 740 741 /* 742 * Read a disk block. 743 * This algorithm described in Bach (p.54). 744 */ 745 int 746 bread(struct vnode *vp, daddr_t blkno, int size, int flags, buf_t **bpp) 747 { 748 buf_t *bp; 749 int error; 750 751 BIOHIST_FUNC(__func__); BIOHIST_CALLED(biohist); 752 753 /* Get buffer for block. */ 754 bp = *bpp = bio_doread(vp, blkno, size, 0); 755 if (bp == NULL) 756 return SET_ERROR(ENOMEM); 757 758 /* Wait for the read to complete, and return result. */ 759 error = biowait(bp); 760 if (error == 0 && (flags & B_MODIFY) != 0) 761 error = fscow_run(bp, true); 762 if (error) { 763 brelse(bp, 0); 764 *bpp = NULL; 765 } 766 767 return error; 768 } 769 770 /* 771 * Read-ahead multiple disk blocks. The first is sync, the rest async. 772 * Trivial modification to the breada algorithm presented in Bach (p.55). 773 */ 774 int 775 breadn(struct vnode *vp, daddr_t blkno, int size, daddr_t *rablks, 776 int *rasizes, int nrablks, int flags, buf_t **bpp) 777 { 778 buf_t *bp; 779 int error, i; 780 781 BIOHIST_FUNC(__func__); BIOHIST_CALLED(biohist); 782 783 bp = *bpp = bio_doread(vp, blkno, size, 0); 784 if (bp == NULL) 785 return SET_ERROR(ENOMEM); 786 787 /* 788 * For each of the read-ahead blocks, start a read, if necessary. 789 */ 790 mutex_enter(&bufcache_lock); 791 for (i = 0; i < nrablks; i++) { 792 /* If it's in the cache, just go on to next one. */ 793 if (incore(vp, rablks[i])) 794 continue; 795 796 /* Get a buffer for the read-ahead block */ 797 mutex_exit(&bufcache_lock); 798 (void) bio_doread(vp, rablks[i], rasizes[i], B_ASYNC); 799 mutex_enter(&bufcache_lock); 800 } 801 mutex_exit(&bufcache_lock); 802 803 /* Otherwise, we had to start a read for it; wait until it's valid. */ 804 error = biowait(bp); 805 if (error == 0 && (flags & B_MODIFY) != 0) 806 error = fscow_run(bp, true); 807 if (error) { 808 brelse(bp, 0); 809 *bpp = NULL; 810 } 811 812 return error; 813 } 814 815 /* 816 * Block write. Described in Bach (p.56) 817 */ 818 int 819 bwrite(buf_t *bp) 820 { 821 int rv, sync, wasdelayed; 822 struct vnode *vp; 823 struct mount *mp; 824 825 BIOHIST_FUNC(__func__); BIOHIST_CALLARGS(biohist, "bp=%#jx", 826 (uintptr_t)bp, 0, 0, 0); 827 828 KASSERT(ISSET(bp->b_cflags, BC_BUSY)); 829 KASSERT(!cv_has_waiters(&bp->b_done)); 830 831 vp = bp->b_vp; 832 833 /* 834 * dholland 20160728 AFAICT vp==NULL must be impossible as it 835 * will crash upon reaching VOP_STRATEGY below... see further 836 * analysis on tech-kern. 837 */ 838 KASSERTMSG(vp != NULL, "bwrite given buffer with null vnode"); 839 840 if (vp != NULL) { 841 KASSERT(bp->b_objlock == vp->v_interlock); 842 if (vp->v_type == VBLK) 843 mp = spec_node_getmountedfs(vp); 844 else 845 mp = vp->v_mount; 846 } else { 847 mp = NULL; 848 } 849 850 if (mp && mp->mnt_wapbl) { 851 if (bp->b_iodone != mp->mnt_wapbl_op->wo_wapbl_biodone) { 852 bdwrite(bp); 853 return 0; 854 } 855 } 856 857 /* 858 * Remember buffer type, to switch on it later. If the write was 859 * synchronous, but the file system was mounted with MNT_ASYNC, 860 * convert it to a delayed write. 861 * XXX note that this relies on delayed tape writes being converted 862 * to async, not sync writes (which is safe, but ugly). 863 */ 864 sync = !ISSET(bp->b_flags, B_ASYNC); 865 if (sync && mp != NULL && ISSET(mp->mnt_flag, MNT_ASYNC)) { 866 bdwrite(bp); 867 return 0; 868 } 869 870 /* 871 * Collect statistics on synchronous and asynchronous writes. 872 * Writes to block devices are charged to their associated 873 * filesystem (if any). 874 */ 875 if (mp != NULL) { 876 if (sync) 877 mp->mnt_stat.f_syncwrites++; 878 else 879 mp->mnt_stat.f_asyncwrites++; 880 } 881 882 /* 883 * Pay for the I/O operation and make sure the buf is on the correct 884 * vnode queue. 885 */ 886 bp->b_error = 0; 887 wasdelayed = ISSET(bp->b_oflags, BO_DELWRI); 888 CLR(bp->b_flags, B_READ); 889 if (wasdelayed) { 890 mutex_enter(&bufcache_lock); 891 mutex_enter(bp->b_objlock); 892 CLR(bp->b_oflags, BO_DONE | BO_DELWRI); 893 reassignbuf(bp, bp->b_vp); 894 /* Wake anyone trying to busy the buffer via vnode's lists. */ 895 cv_broadcast(&bp->b_busy); 896 mutex_exit(&bufcache_lock); 897 } else { 898 curlwp->l_ru.ru_oublock++; 899 mutex_enter(bp->b_objlock); 900 CLR(bp->b_oflags, BO_DONE | BO_DELWRI); 901 } 902 if (vp != NULL) 903 vp->v_numoutput++; 904 mutex_exit(bp->b_objlock); 905 906 /* Initiate disk write. */ 907 if (sync) 908 BIO_SETPRIO(bp, BPRIO_TIMECRITICAL); 909 else 910 BIO_SETPRIO(bp, BPRIO_TIMELIMITED); 911 912 VOP_STRATEGY(vp, bp); 913 914 if (sync) { 915 /* If I/O was synchronous, wait for it to complete. */ 916 rv = biowait(bp); 917 918 /* Release the buffer. */ 919 brelse(bp, 0); 920 921 return rv; 922 } else { 923 return 0; 924 } 925 } 926 927 int 928 vn_bwrite(void *v) 929 { 930 struct vop_bwrite_args *ap = v; 931 932 return bwrite(ap->a_bp); 933 } 934 935 /* 936 * Delayed write. 937 * 938 * The buffer is marked dirty, but is not queued for I/O. 939 * This routine should be used when the buffer is expected 940 * to be modified again soon, typically a small write that 941 * partially fills a buffer. 942 * 943 * NB: magnetic tapes cannot be delayed; they must be 944 * written in the order that the writes are requested. 945 * 946 * Described in Leffler, et al. (pp. 208-213). 947 */ 948 void 949 bdwrite(buf_t *bp) 950 { 951 952 BIOHIST_FUNC(__func__); BIOHIST_CALLARGS(biohist, "bp=%#jx", 953 (uintptr_t)bp, 0, 0, 0); 954 955 KASSERT(bp->b_vp == NULL || bp->b_vp->v_tag != VT_UFS || 956 bp->b_vp->v_type == VBLK || ISSET(bp->b_flags, B_COWDONE)); 957 KASSERT(ISSET(bp->b_cflags, BC_BUSY)); 958 KASSERT(!cv_has_waiters(&bp->b_done)); 959 960 /* If this is a tape block, write the block now. */ 961 if (bdev_type(bp->b_dev) == D_TAPE) { 962 bawrite(bp); 963 return; 964 } 965 966 if (wapbl_vphaswapbl(bp->b_vp)) { 967 struct mount *mp = wapbl_vptomp(bp->b_vp); 968 969 if (bp->b_iodone != mp->mnt_wapbl_op->wo_wapbl_biodone) { 970 WAPBL_ADD_BUF(mp, bp); 971 } 972 } 973 974 /* 975 * If the block hasn't been seen before: 976 * (1) Mark it as having been seen, 977 * (2) Charge for the write, 978 * (3) Make sure it's on its vnode's correct block list. 979 */ 980 KASSERT(bp->b_vp == NULL || bp->b_objlock == bp->b_vp->v_interlock); 981 982 if (!ISSET(bp->b_oflags, BO_DELWRI)) { 983 mutex_enter(&bufcache_lock); 984 mutex_enter(bp->b_objlock); 985 SET(bp->b_oflags, BO_DELWRI); 986 curlwp->l_ru.ru_oublock++; 987 reassignbuf(bp, bp->b_vp); 988 /* Wake anyone trying to busy the buffer via vnode's lists. */ 989 cv_broadcast(&bp->b_busy); 990 mutex_exit(&bufcache_lock); 991 } else { 992 mutex_enter(bp->b_objlock); 993 } 994 /* Otherwise, the "write" is done, so mark and release the buffer. */ 995 CLR(bp->b_oflags, BO_DONE); 996 mutex_exit(bp->b_objlock); 997 998 brelse(bp, 0); 999 } 1000 1001 /* 1002 * Asynchronous block write; just an asynchronous bwrite(). 1003 */ 1004 void 1005 bawrite(buf_t *bp) 1006 { 1007 1008 KASSERT(ISSET(bp->b_cflags, BC_BUSY)); 1009 KASSERT(bp->b_vp != NULL); 1010 1011 SET(bp->b_flags, B_ASYNC); 1012 VOP_BWRITE(bp->b_vp, bp); 1013 } 1014 1015 /* 1016 * Release a buffer on to the free lists. 1017 * Described in Bach (p. 46). 1018 */ 1019 void 1020 brelsel(buf_t *bp, int set) 1021 { 1022 struct bqueue *bufq; 1023 struct vnode *vp; 1024 1025 SDT_PROBE2(io, kernel, , brelse, bp, set); 1026 1027 KASSERT(bp != NULL); 1028 KASSERT(mutex_owned(&bufcache_lock)); 1029 KASSERT(!cv_has_waiters(&bp->b_done)); 1030 1031 SET(bp->b_cflags, set); 1032 1033 KASSERT(ISSET(bp->b_cflags, BC_BUSY)); 1034 KASSERT(bp->b_iodone == NULL); 1035 1036 /* Wake up any processes waiting for any buffer to become free. */ 1037 cv_signal(&needbuffer_cv); 1038 1039 /* Wake up any proceeses waiting for _this_ buffer to become free */ 1040 if (ISSET(bp->b_cflags, BC_WANTED)) 1041 CLR(bp->b_cflags, BC_WANTED|BC_AGE); 1042 1043 /* If it's clean clear the copy-on-write flag. */ 1044 if (ISSET(bp->b_flags, B_COWDONE)) { 1045 mutex_enter(bp->b_objlock); 1046 if (!ISSET(bp->b_oflags, BO_DELWRI)) 1047 CLR(bp->b_flags, B_COWDONE); 1048 mutex_exit(bp->b_objlock); 1049 } 1050 1051 /* 1052 * Determine which queue the buffer should be on, then put it there. 1053 */ 1054 1055 /* If it's locked, don't report an error; try again later. */ 1056 if (ISSET(bp->b_flags, B_LOCKED)) 1057 bp->b_error = 0; 1058 1059 /* If it's not cacheable, or an error, mark it invalid. */ 1060 if (ISSET(bp->b_cflags, BC_NOCACHE) || bp->b_error != 0) 1061 SET(bp->b_cflags, BC_INVAL); 1062 1063 if (ISSET(bp->b_cflags, BC_VFLUSH)) { 1064 /* 1065 * This is a delayed write buffer that was just flushed to 1066 * disk. It is still on the LRU queue. If it's become 1067 * invalid, then we need to move it to a different queue; 1068 * otherwise leave it in its current position. 1069 */ 1070 CLR(bp->b_cflags, BC_VFLUSH); 1071 if (!ISSET(bp->b_cflags, BC_INVAL|BC_AGE) && 1072 !ISSET(bp->b_flags, B_LOCKED) && bp->b_error == 0) { 1073 KDASSERT(checkfreelist(bp, &bufqueues[BQ_LRU], 1)); 1074 goto already_queued; 1075 } else { 1076 bremfree(bp); 1077 } 1078 } 1079 1080 KDASSERT(checkfreelist(bp, &bufqueues[BQ_AGE], 0)); 1081 KDASSERT(checkfreelist(bp, &bufqueues[BQ_LRU], 0)); 1082 KDASSERT(checkfreelist(bp, &bufqueues[BQ_LOCKED], 0)); 1083 1084 if ((bp->b_bufsize <= 0) || ISSET(bp->b_cflags, BC_INVAL)) { 1085 /* 1086 * If it's invalid or empty, dissociate it from its vnode 1087 * and put on the head of the appropriate queue. 1088 */ 1089 if (ISSET(bp->b_flags, B_LOCKED)) { 1090 if (wapbl_vphaswapbl(vp = bp->b_vp)) { 1091 struct mount *mp = wapbl_vptomp(vp); 1092 1093 KASSERT(bp->b_iodone != 1094 mp->mnt_wapbl_op->wo_wapbl_biodone); 1095 WAPBL_REMOVE_BUF(mp, bp); 1096 } 1097 } 1098 1099 mutex_enter(bp->b_objlock); 1100 CLR(bp->b_oflags, BO_DONE|BO_DELWRI); 1101 if ((vp = bp->b_vp) != NULL) { 1102 KASSERT(bp->b_objlock == vp->v_interlock); 1103 reassignbuf(bp, bp->b_vp); 1104 brelvp(bp); 1105 mutex_exit(vp->v_interlock); 1106 } else { 1107 KASSERT(bp->b_objlock == &buffer_lock); 1108 mutex_exit(bp->b_objlock); 1109 } 1110 /* We want to dispose of the buffer, so wake everybody. */ 1111 cv_broadcast(&bp->b_busy); 1112 if (bp->b_bufsize <= 0) 1113 /* no data */ 1114 goto already_queued; 1115 else 1116 /* invalid data */ 1117 bufq = &bufqueues[BQ_AGE]; 1118 binsheadfree(bp, bufq); 1119 } else { 1120 /* 1121 * It has valid data. Put it on the end of the appropriate 1122 * queue, so that it'll stick around for as long as possible. 1123 * If buf is AGE, but has dependencies, must put it on last 1124 * bufqueue to be scanned, ie LRU. This protects against the 1125 * livelock where BQ_AGE only has buffers with dependencies, 1126 * and we thus never get to the dependent buffers in BQ_LRU. 1127 */ 1128 if (ISSET(bp->b_flags, B_LOCKED)) { 1129 /* locked in core */ 1130 bufq = &bufqueues[BQ_LOCKED]; 1131 } else if (!ISSET(bp->b_cflags, BC_AGE)) { 1132 /* valid data */ 1133 bufq = &bufqueues[BQ_LRU]; 1134 } else { 1135 /* stale but valid data */ 1136 bufq = &bufqueues[BQ_AGE]; 1137 } 1138 binstailfree(bp, bufq); 1139 } 1140 already_queued: 1141 /* Unlock the buffer. */ 1142 CLR(bp->b_cflags, BC_AGE|BC_BUSY|BC_NOCACHE); 1143 CLR(bp->b_flags, B_ASYNC); 1144 1145 /* 1146 * Wake only the highest priority waiter on the lock, in order to 1147 * prevent a thundering herd: many LWPs simultaneously awakening and 1148 * competing for the buffer's lock. Testing in 2019 revealed this 1149 * to reduce contention on bufcache_lock tenfold during a kernel 1150 * compile. Here and elsewhere, when the buffer is changing 1151 * identity, being disposed of, or moving from one list to another, 1152 * we wake all lock requestors. 1153 */ 1154 if (bp->b_bufsize <= 0) { 1155 cv_broadcast(&bp->b_busy); 1156 buf_destroy(bp); 1157 #ifdef DEBUG 1158 memset((char *)bp, 0, sizeof(*bp)); 1159 #endif 1160 pool_cache_put(buf_cache, bp); 1161 } else 1162 cv_signal(&bp->b_busy); 1163 } 1164 1165 void 1166 brelse(buf_t *bp, int set) 1167 { 1168 1169 mutex_enter(&bufcache_lock); 1170 brelsel(bp, set); 1171 mutex_exit(&bufcache_lock); 1172 } 1173 1174 /* 1175 * Determine if a block is in the cache. 1176 * Just look on what would be its hash chain. If it's there, return 1177 * a pointer to it, unless it's marked invalid. If it's marked invalid, 1178 * we normally don't return the buffer, unless the caller explicitly 1179 * wants us to. 1180 */ 1181 buf_t * 1182 incore(struct vnode *vp, daddr_t blkno) 1183 { 1184 buf_t *bp; 1185 1186 KASSERT(mutex_owned(&bufcache_lock)); 1187 1188 /* Search hash chain */ 1189 LIST_FOREACH(bp, BUFHASH(vp, blkno), b_hash) { 1190 if (bp->b_lblkno == blkno && bp->b_vp == vp && 1191 !ISSET(bp->b_cflags, BC_INVAL)) { 1192 KASSERT(bp->b_objlock == vp->v_interlock); 1193 return (bp); 1194 } 1195 } 1196 1197 return NULL; 1198 } 1199 1200 /* 1201 * Get a block of requested size that is associated with 1202 * a given vnode and block offset. If it is found in the 1203 * block cache, mark it as having been found, make it busy 1204 * and return it. Otherwise, return an empty block of the 1205 * correct size. It is up to the caller to insure that the 1206 * cached blocks be of the correct size. 1207 */ 1208 buf_t * 1209 getblk(struct vnode *vp, daddr_t blkno, int size, int slpflag, int slptimeo) 1210 { 1211 int err, preserve; 1212 buf_t *bp; 1213 1214 mutex_enter(&bufcache_lock); 1215 SDT_PROBE3(io, kernel, , getblk__start, vp, blkno, size); 1216 loop: 1217 bp = incore(vp, blkno); 1218 if (bp != NULL) { 1219 err = bbusy(bp, ((slpflag & PCATCH) != 0), slptimeo, NULL); 1220 if (err != 0) { 1221 if (err == EPASSTHROUGH) 1222 goto loop; 1223 mutex_exit(&bufcache_lock); 1224 SDT_PROBE4(io, kernel, , getblk__done, 1225 vp, blkno, size, NULL); 1226 return NULL; 1227 } 1228 KASSERT(!cv_has_waiters(&bp->b_done)); 1229 #ifdef DIAGNOSTIC 1230 if (ISSET(bp->b_oflags, BO_DONE|BO_DELWRI) && 1231 bp->b_bcount < size && vp->v_type != VBLK) 1232 panic("getblk: block size invariant failed"); 1233 #endif 1234 bremfree(bp); 1235 preserve = 1; 1236 } else { 1237 if ((bp = getnewbuf(slpflag, slptimeo, 0)) == NULL) 1238 goto loop; 1239 1240 if (incore(vp, blkno) != NULL) { 1241 /* The block has come into memory in the meantime. */ 1242 brelsel(bp, 0); 1243 goto loop; 1244 } 1245 1246 LIST_INSERT_HEAD(BUFHASH(vp, blkno), bp, b_hash); 1247 bp->b_blkno = bp->b_lblkno = bp->b_rawblkno = blkno; 1248 mutex_enter(vp->v_interlock); 1249 bgetvp(vp, bp); 1250 mutex_exit(vp->v_interlock); 1251 preserve = 0; 1252 } 1253 mutex_exit(&bufcache_lock); 1254 1255 /* 1256 * LFS can't track total size of B_LOCKED buffer (locked_queue_bytes) 1257 * if we re-size buffers here. 1258 */ 1259 if (ISSET(bp->b_flags, B_LOCKED)) { 1260 KASSERT(bp->b_bufsize >= size); 1261 } else { 1262 if (allocbuf(bp, size, preserve)) { 1263 mutex_enter(&bufcache_lock); 1264 LIST_REMOVE(bp, b_hash); 1265 brelsel(bp, BC_INVAL); 1266 mutex_exit(&bufcache_lock); 1267 SDT_PROBE4(io, kernel, , getblk__done, 1268 vp, blkno, size, NULL); 1269 return NULL; 1270 } 1271 } 1272 BIO_SETPRIO(bp, BPRIO_DEFAULT); 1273 SDT_PROBE4(io, kernel, , getblk__done, vp, blkno, size, bp); 1274 return bp; 1275 } 1276 1277 /* 1278 * Get an empty, disassociated buffer of given size. 1279 */ 1280 buf_t * 1281 geteblk(int size) 1282 { 1283 buf_t *bp; 1284 int error __diagused; 1285 1286 mutex_enter(&bufcache_lock); 1287 while ((bp = getnewbuf(0, 0, 0)) == NULL) 1288 continue; 1289 1290 SET(bp->b_cflags, BC_INVAL); 1291 LIST_INSERT_HEAD(&invalhash, bp, b_hash); 1292 mutex_exit(&bufcache_lock); 1293 BIO_SETPRIO(bp, BPRIO_DEFAULT); 1294 error = allocbuf(bp, size, 0); 1295 KASSERT(error == 0); 1296 return bp; 1297 } 1298 1299 /* 1300 * Expand or contract the actual memory allocated to a buffer. 1301 * 1302 * If the buffer shrinks, data is lost, so it's up to the 1303 * caller to have written it out *first*; this routine will not 1304 * start a write. If the buffer grows, it's the callers 1305 * responsibility to fill out the buffer's additional contents. 1306 */ 1307 int 1308 allocbuf(buf_t *bp, int size, int preserve) 1309 { 1310 void *addr; 1311 vsize_t oldsize, desired_size; 1312 int oldcount; 1313 int delta; 1314 1315 desired_size = buf_roundsize(size); 1316 if (desired_size > MAXBSIZE) 1317 printf("allocbuf: buffer larger than MAXBSIZE requested"); 1318 1319 oldcount = bp->b_bcount; 1320 1321 bp->b_bcount = size; 1322 1323 oldsize = bp->b_bufsize; 1324 if (oldsize == desired_size) { 1325 /* 1326 * Do not short cut the WAPBL resize, as the buffer length 1327 * could still have changed and this would corrupt the 1328 * tracking of the transaction length. 1329 */ 1330 goto out; 1331 } 1332 1333 /* 1334 * If we want a buffer of a different size, re-allocate the 1335 * buffer's memory; copy old content only if needed. 1336 */ 1337 addr = buf_alloc(desired_size); 1338 if (addr == NULL) 1339 return SET_ERROR(ENOMEM); 1340 if (preserve) 1341 memcpy(addr, bp->b_data, MIN(oldsize,desired_size)); 1342 if (bp->b_data != NULL) 1343 buf_mrelease(bp->b_data, oldsize); 1344 bp->b_data = addr; 1345 bp->b_bufsize = desired_size; 1346 1347 /* 1348 * Update overall buffer memory counter (protected by bufcache_lock) 1349 */ 1350 delta = (long)desired_size - (long)oldsize; 1351 1352 mutex_enter(&bufcache_lock); 1353 if ((bufmem += delta) > bufmem_hiwater) { 1354 /* 1355 * Need to trim overall memory usage. 1356 */ 1357 while (buf_canrelease()) { 1358 if (preempt_needed()) { 1359 mutex_exit(&bufcache_lock); 1360 preempt(); 1361 mutex_enter(&bufcache_lock); 1362 } 1363 if (buf_trim() == 0) 1364 break; 1365 } 1366 } 1367 mutex_exit(&bufcache_lock); 1368 1369 out: 1370 if (wapbl_vphaswapbl(bp->b_vp)) { 1371 WAPBL_RESIZE_BUF(wapbl_vptomp(bp->b_vp), bp, 1372 oldsize, oldcount); 1373 } 1374 1375 return 0; 1376 } 1377 1378 /* 1379 * Find a buffer which is available for use. 1380 * Select something from a free list. 1381 * Preference is to AGE list, then LRU list. 1382 * 1383 * Called with the buffer queues locked. 1384 * Return buffer locked. 1385 */ 1386 static buf_t * 1387 getnewbuf(int slpflag, int slptimeo, int from_bufq) 1388 { 1389 buf_t *bp; 1390 struct vnode *vp; 1391 struct mount *transmp = NULL; 1392 1393 SDT_PROBE0(io, kernel, , getnewbuf__start); 1394 1395 start: 1396 KASSERT(mutex_owned(&bufcache_lock)); 1397 1398 /* 1399 * Get a new buffer from the pool. 1400 */ 1401 if (!from_bufq && buf_lotsfree()) { 1402 mutex_exit(&bufcache_lock); 1403 bp = pool_cache_get(buf_cache, PR_NOWAIT); 1404 if (bp != NULL) { 1405 memset((char *)bp, 0, sizeof(*bp)); 1406 buf_init(bp); 1407 SET(bp->b_cflags, BC_BUSY); /* mark buffer busy */ 1408 mutex_enter(&bufcache_lock); 1409 #if defined(DIAGNOSTIC) 1410 bp->b_freelistindex = -1; 1411 #endif /* defined(DIAGNOSTIC) */ 1412 SDT_PROBE1(io, kernel, , getnewbuf__done, bp); 1413 return bp; 1414 } 1415 mutex_enter(&bufcache_lock); 1416 } 1417 1418 KASSERT(mutex_owned(&bufcache_lock)); 1419 if ((bp = TAILQ_FIRST(&bufqueues[BQ_AGE].bq_queue)) != NULL) { 1420 KASSERT(!ISSET(bp->b_oflags, BO_DELWRI)); 1421 } else { 1422 TAILQ_FOREACH(bp, &bufqueues[BQ_LRU].bq_queue, b_freelist) { 1423 if (ISSET(bp->b_cflags, BC_VFLUSH) || 1424 !ISSET(bp->b_oflags, BO_DELWRI)) 1425 break; 1426 if (fstrans_start_nowait(bp->b_vp->v_mount) == 0) { 1427 KASSERT(transmp == NULL); 1428 transmp = bp->b_vp->v_mount; 1429 break; 1430 } 1431 } 1432 } 1433 if (bp != NULL) { 1434 KASSERT(!ISSET(bp->b_cflags, BC_BUSY) || 1435 ISSET(bp->b_cflags, BC_VFLUSH)); 1436 bremfree(bp); 1437 1438 /* Buffer is no longer on free lists. */ 1439 SET(bp->b_cflags, BC_BUSY); 1440 1441 /* Wake anyone trying to lock the old identity. */ 1442 cv_broadcast(&bp->b_busy); 1443 } else { 1444 /* 1445 * XXX: !from_bufq should be removed. 1446 */ 1447 if (!from_bufq || curlwp != uvm.pagedaemon_lwp) { 1448 /* wait for a free buffer of any kind */ 1449 if ((slpflag & PCATCH) != 0) 1450 (void)cv_timedwait_sig(&needbuffer_cv, 1451 &bufcache_lock, slptimeo); 1452 else 1453 (void)cv_timedwait(&needbuffer_cv, 1454 &bufcache_lock, slptimeo); 1455 } 1456 SDT_PROBE1(io, kernel, , getnewbuf__done, NULL); 1457 return NULL; 1458 } 1459 1460 #ifdef DIAGNOSTIC 1461 if (bp->b_bufsize <= 0) 1462 panic("buffer %p: on queue but empty", bp); 1463 #endif 1464 1465 if (ISSET(bp->b_cflags, BC_VFLUSH)) { 1466 /* 1467 * This is a delayed write buffer being flushed to disk. Make 1468 * sure it gets aged out of the queue when it's finished, and 1469 * leave it off the LRU queue. 1470 */ 1471 CLR(bp->b_cflags, BC_VFLUSH); 1472 SET(bp->b_cflags, BC_AGE); 1473 goto start; 1474 } 1475 1476 KASSERT(ISSET(bp->b_cflags, BC_BUSY)); 1477 KASSERT(!cv_has_waiters(&bp->b_done)); 1478 1479 /* 1480 * If buffer was a delayed write, start it and return NULL 1481 * (since we might sleep while starting the write). 1482 */ 1483 if (ISSET(bp->b_oflags, BO_DELWRI)) { 1484 /* 1485 * This buffer has gone through the LRU, so make sure it gets 1486 * reused ASAP. 1487 */ 1488 SET(bp->b_cflags, BC_AGE); 1489 mutex_exit(&bufcache_lock); 1490 bawrite(bp); 1491 KASSERT(transmp != NULL); 1492 fstrans_done(transmp); 1493 mutex_enter(&bufcache_lock); 1494 SDT_PROBE1(io, kernel, , getnewbuf__done, NULL); 1495 return NULL; 1496 } 1497 1498 KASSERT(transmp == NULL); 1499 1500 vp = bp->b_vp; 1501 1502 /* clear out various other fields */ 1503 bp->b_cflags = BC_BUSY; 1504 bp->b_oflags = 0; 1505 bp->b_flags = 0; 1506 bp->b_dev = NODEV; 1507 bp->b_blkno = 0; 1508 bp->b_lblkno = 0; 1509 bp->b_rawblkno = 0; 1510 bp->b_iodone = 0; 1511 bp->b_error = 0; 1512 bp->b_resid = 0; 1513 bp->b_bcount = 0; 1514 1515 LIST_REMOVE(bp, b_hash); 1516 1517 /* Disassociate us from our vnode, if we had one... */ 1518 if (vp != NULL) { 1519 mutex_enter(vp->v_interlock); 1520 brelvp(bp); 1521 mutex_exit(vp->v_interlock); 1522 } 1523 1524 SDT_PROBE1(io, kernel, , getnewbuf__done, bp); 1525 return bp; 1526 } 1527 1528 /* 1529 * Invalidate the specified buffer if it exists. 1530 */ 1531 void 1532 binvalbuf(struct vnode *vp, daddr_t blkno) 1533 { 1534 buf_t *bp; 1535 int err; 1536 1537 mutex_enter(&bufcache_lock); 1538 1539 loop: 1540 bp = incore(vp, blkno); 1541 if (bp != NULL) { 1542 err = bbusy(bp, 0, 0, NULL); 1543 if (err == EPASSTHROUGH) 1544 goto loop; 1545 bremfree(bp); 1546 if (ISSET(bp->b_oflags, BO_DELWRI)) { 1547 SET(bp->b_cflags, BC_NOCACHE); 1548 mutex_exit(&bufcache_lock); 1549 bwrite(bp); 1550 } else { 1551 brelsel(bp, BC_INVAL); 1552 mutex_exit(&bufcache_lock); 1553 } 1554 } else 1555 mutex_exit(&bufcache_lock); 1556 } 1557 1558 /* 1559 * Attempt to free an aged buffer off the queues. 1560 * Called with queue lock held. 1561 * Returns the amount of buffer memory freed. 1562 */ 1563 static int 1564 buf_trim(void) 1565 { 1566 buf_t *bp; 1567 long size; 1568 1569 KASSERT(mutex_owned(&bufcache_lock)); 1570 1571 /* Instruct getnewbuf() to get buffers off the queues */ 1572 if ((bp = getnewbuf(PCATCH, 1, 1)) == NULL) 1573 return 0; 1574 1575 KASSERT((bp->b_cflags & BC_WANTED) == 0); 1576 size = bp->b_bufsize; 1577 bufmem -= size; 1578 if (size > 0) { 1579 buf_mrelease(bp->b_data, size); 1580 bp->b_bcount = bp->b_bufsize = 0; 1581 } 1582 /* brelse() will return the buffer to the global buffer pool */ 1583 brelsel(bp, 0); 1584 return size; 1585 } 1586 1587 int 1588 buf_drain(int n) 1589 { 1590 int size = 0, sz; 1591 1592 KASSERT(mutex_owned(&bufcache_lock)); 1593 1594 while (size < n && bufmem > bufmem_lowater) { 1595 sz = buf_trim(); 1596 if (sz <= 0) 1597 break; 1598 size += sz; 1599 } 1600 1601 return size; 1602 } 1603 1604 /* 1605 * Wait for operations on the buffer to complete. 1606 * When they do, extract and return the I/O's error value. 1607 */ 1608 int 1609 biowait(buf_t *bp) 1610 { 1611 1612 BIOHIST_FUNC(__func__); 1613 1614 KASSERT(ISSET(bp->b_cflags, BC_BUSY)); 1615 1616 SDT_PROBE1(io, kernel, , wait__start, bp); 1617 1618 mutex_enter(bp->b_objlock); 1619 1620 BIOHIST_CALLARGS(biohist, "bp=%#jx, oflags=0x%jx, ret_addr=%#jx", 1621 (uintptr_t)bp, bp->b_oflags, 1622 (uintptr_t)__builtin_return_address(0), 0); 1623 1624 while (!ISSET(bp->b_oflags, BO_DONE | BO_DELWRI)) { 1625 BIOHIST_LOG(biohist, "waiting bp=%#jx", 1626 (uintptr_t)bp, 0, 0, 0); 1627 cv_wait(&bp->b_done, bp->b_objlock); 1628 } 1629 mutex_exit(bp->b_objlock); 1630 1631 SDT_PROBE1(io, kernel, , wait__done, bp); 1632 1633 BIOHIST_LOG(biohist, "return %jd", bp->b_error, 0, 0, 0); 1634 1635 return bp->b_error; 1636 } 1637 1638 /* 1639 * Mark I/O complete on a buffer. 1640 * 1641 * If a callback has been requested, e.g. the pageout 1642 * daemon, do so. Otherwise, awaken waiting processes. 1643 * 1644 * [ Leffler, et al., says on p.247: 1645 * "This routine wakes up the blocked process, frees the buffer 1646 * for an asynchronous write, or, for a request by the pagedaemon 1647 * process, invokes a procedure specified in the buffer structure" ] 1648 * 1649 * In real life, the pagedaemon (or other system processes) wants 1650 * to do async stuff too, and doesn't want the buffer brelse()'d. 1651 * (for swap pager, that puts swap buffers on the free lists (!!!), 1652 * for the vn device, that puts allocated buffers on the free lists!) 1653 */ 1654 void 1655 biodone(buf_t *bp) 1656 { 1657 int s; 1658 1659 BIOHIST_FUNC(__func__); 1660 1661 KASSERT(!ISSET(bp->b_oflags, BO_DONE)); 1662 1663 if (cpu_intr_p()) { 1664 /* From interrupt mode: defer to a soft interrupt. */ 1665 s = splvm(); 1666 TAILQ_INSERT_TAIL(&curcpu()->ci_data.cpu_biodone, bp, b_actq); 1667 1668 BIOHIST_CALLARGS(biohist, "bp=%#jx, softint scheduled", 1669 (uintptr_t)bp, 0, 0, 0); 1670 softint_schedule(biodone_sih); 1671 splx(s); 1672 } else { 1673 /* Process now - the buffer may be freed soon. */ 1674 biodone2(bp); 1675 } 1676 } 1677 1678 SDT_PROBE_DEFINE1(io, kernel, , done, "struct buf *"/*bp*/); 1679 1680 static void 1681 biodone2(buf_t *bp) 1682 { 1683 void (*callout)(buf_t *); 1684 1685 SDT_PROBE1(io, kernel, ,done, bp); 1686 1687 BIOHIST_FUNC(__func__); 1688 BIOHIST_CALLARGS(biohist, "bp=%#jx", (uintptr_t)bp, 0, 0, 0); 1689 1690 mutex_enter(bp->b_objlock); 1691 /* Note that the transfer is done. */ 1692 if (ISSET(bp->b_oflags, BO_DONE)) 1693 panic("biodone2 already"); 1694 CLR(bp->b_flags, B_COWDONE); 1695 SET(bp->b_oflags, BO_DONE); 1696 BIO_SETPRIO(bp, BPRIO_DEFAULT); 1697 1698 /* Wake up waiting writers. */ 1699 if (!ISSET(bp->b_flags, B_READ)) 1700 vwakeup(bp); 1701 1702 if ((callout = bp->b_iodone) != NULL) { 1703 BIOHIST_LOG(biohist, "callout %#jx", (uintptr_t)callout, 1704 0, 0, 0); 1705 1706 /* Note callout done, then call out. */ 1707 KASSERT(!cv_has_waiters(&bp->b_done)); 1708 bp->b_iodone = NULL; 1709 mutex_exit(bp->b_objlock); 1710 (*callout)(bp); 1711 } else if (ISSET(bp->b_flags, B_ASYNC)) { 1712 /* If async, release. */ 1713 BIOHIST_LOG(biohist, "async", 0, 0, 0, 0); 1714 KASSERT(!cv_has_waiters(&bp->b_done)); 1715 mutex_exit(bp->b_objlock); 1716 brelse(bp, 0); 1717 } else { 1718 /* Otherwise just wake up waiters in biowait(). */ 1719 BIOHIST_LOG(biohist, "wake-up", 0, 0, 0, 0); 1720 cv_broadcast(&bp->b_done); 1721 mutex_exit(bp->b_objlock); 1722 } 1723 } 1724 1725 static void 1726 biointr(void *cookie) 1727 { 1728 struct cpu_info *ci; 1729 buf_t *bp; 1730 int s; 1731 1732 BIOHIST_FUNC(__func__); BIOHIST_CALLED(biohist); 1733 1734 ci = curcpu(); 1735 1736 s = splvm(); 1737 while (!TAILQ_EMPTY(&ci->ci_data.cpu_biodone)) { 1738 KASSERT(curcpu() == ci); 1739 1740 bp = TAILQ_FIRST(&ci->ci_data.cpu_biodone); 1741 TAILQ_REMOVE(&ci->ci_data.cpu_biodone, bp, b_actq); 1742 splx(s); 1743 1744 BIOHIST_LOG(biohist, "bp=%#jx", (uintptr_t)bp, 0, 0, 0); 1745 biodone2(bp); 1746 1747 s = splvm(); 1748 } 1749 splx(s); 1750 } 1751 1752 static void 1753 sysctl_fillbuf(const buf_t *i, struct buf_sysctl *o) 1754 { 1755 const bool allowaddr = get_expose_address(curproc); 1756 1757 memset(o, 0, sizeof(*o)); 1758 1759 o->b_flags = i->b_flags | i->b_cflags | i->b_oflags; 1760 o->b_error = i->b_error; 1761 o->b_prio = i->b_prio; 1762 o->b_dev = i->b_dev; 1763 o->b_bufsize = i->b_bufsize; 1764 o->b_bcount = i->b_bcount; 1765 o->b_resid = i->b_resid; 1766 COND_SET_VALUE(o->b_addr, PTRTOUINT64(i->b_data), allowaddr); 1767 o->b_blkno = i->b_blkno; 1768 o->b_rawblkno = i->b_rawblkno; 1769 COND_SET_VALUE(o->b_iodone, PTRTOUINT64(i->b_iodone), allowaddr); 1770 COND_SET_VALUE(o->b_proc, PTRTOUINT64(i->b_proc), allowaddr); 1771 COND_SET_VALUE(o->b_vp, PTRTOUINT64(i->b_vp), allowaddr); 1772 COND_SET_VALUE(o->b_saveaddr, PTRTOUINT64(i->b_saveaddr), allowaddr); 1773 o->b_lblkno = i->b_lblkno; 1774 } 1775 1776 static int 1777 sysctl_dobuf(SYSCTLFN_ARGS) 1778 { 1779 buf_t *bp; 1780 struct buf_sysctl bs; 1781 struct bqueue *bq; 1782 char *dp; 1783 u_int i, op, arg; 1784 size_t len, needed, elem_size, out_size; 1785 int error, elem_count, retries; 1786 1787 if (namelen == 1 && name[0] == CTL_QUERY) 1788 return sysctl_query(SYSCTLFN_CALL(rnode)); 1789 1790 if (namelen != 4) 1791 return SET_ERROR(EINVAL); 1792 1793 retries = 100; 1794 retry: 1795 dp = oldp; 1796 len = (oldp != NULL) ? *oldlenp : 0; 1797 op = name[0]; 1798 arg = name[1]; 1799 elem_size = name[2]; 1800 elem_count = name[3]; 1801 out_size = MIN(sizeof(bs), elem_size); 1802 1803 /* 1804 * at the moment, these are just "placeholders" to make the 1805 * API for retrieving kern.buf data more extensible in the 1806 * future. 1807 * 1808 * XXX kern.buf currently has "netbsd32" issues. hopefully 1809 * these will be resolved at a later point. 1810 */ 1811 if (op != KERN_BUF_ALL || arg != KERN_BUF_ALL || 1812 elem_size < 1 || elem_count < 0) 1813 return SET_ERROR(EINVAL); 1814 1815 if (oldp == NULL) { 1816 /* count only, don't run through the buffer queues */ 1817 needed = pool_cache_nget(buf_cache) - 1818 pool_cache_nput(buf_cache); 1819 *oldlenp = (needed + KERN_BUFSLOP) * elem_size; 1820 1821 return 0; 1822 } 1823 1824 error = 0; 1825 needed = 0; 1826 sysctl_unlock(); 1827 mutex_enter(&bufcache_lock); 1828 for (i = 0; i < BQUEUES; i++) { 1829 bq = &bufqueues[i]; 1830 TAILQ_FOREACH(bp, &bq->bq_queue, b_freelist) { 1831 bq->bq_marker = bp; 1832 if (len >= elem_size && elem_count > 0) { 1833 sysctl_fillbuf(bp, &bs); 1834 mutex_exit(&bufcache_lock); 1835 error = copyout(&bs, dp, out_size); 1836 mutex_enter(&bufcache_lock); 1837 if (error) 1838 break; 1839 if (bq->bq_marker != bp) { 1840 /* 1841 * This sysctl node is only for 1842 * statistics. Retry; if the 1843 * queue keeps changing, then 1844 * bail out. 1845 */ 1846 if (retries-- == 0) { 1847 error = SET_ERROR(EAGAIN); 1848 break; 1849 } 1850 mutex_exit(&bufcache_lock); 1851 sysctl_relock(); 1852 goto retry; 1853 } 1854 dp += elem_size; 1855 len -= elem_size; 1856 } 1857 needed += elem_size; 1858 if (elem_count > 0 && elem_count != INT_MAX) 1859 elem_count--; 1860 } 1861 if (error != 0) 1862 break; 1863 } 1864 mutex_exit(&bufcache_lock); 1865 sysctl_relock(); 1866 1867 *oldlenp = needed; 1868 1869 return error; 1870 } 1871 1872 static int 1873 sysctl_bufvm_update(SYSCTLFN_ARGS) 1874 { 1875 int error, rv; 1876 struct sysctlnode node; 1877 unsigned int temp_bufcache; 1878 unsigned long temp_water; 1879 1880 /* Take a copy of the supplied node and its data */ 1881 node = *rnode; 1882 if (node.sysctl_data == &bufcache) { 1883 node.sysctl_data = &temp_bufcache; 1884 temp_bufcache = *(unsigned int *)rnode->sysctl_data; 1885 } else { 1886 node.sysctl_data = &temp_water; 1887 temp_water = *(unsigned long *)rnode->sysctl_data; 1888 } 1889 1890 /* Update the copy */ 1891 error = sysctl_lookup(SYSCTLFN_CALL(&node)); 1892 if (error || newp == NULL) 1893 return error; 1894 1895 if (rnode->sysctl_data == &bufcache) { 1896 if (temp_bufcache > 100) 1897 return SET_ERROR(EINVAL); 1898 bufcache = temp_bufcache; 1899 buf_setwm(); 1900 } else if (rnode->sysctl_data == &bufmem_lowater) { 1901 if (bufmem_hiwater - temp_water < 16) 1902 return SET_ERROR(EINVAL); 1903 bufmem_lowater = temp_water; 1904 } else if (rnode->sysctl_data == &bufmem_hiwater) { 1905 if (temp_water - bufmem_lowater < 16) 1906 return SET_ERROR(EINVAL); 1907 bufmem_hiwater = temp_water; 1908 } else 1909 return SET_ERROR(EINVAL); 1910 1911 /* Drain until below new high water mark */ 1912 sysctl_unlock(); 1913 mutex_enter(&bufcache_lock); 1914 while (bufmem > bufmem_hiwater) { 1915 rv = buf_drain((bufmem - bufmem_hiwater) / (2 * 1024)); 1916 if (rv <= 0) 1917 break; 1918 } 1919 mutex_exit(&bufcache_lock); 1920 sysctl_relock(); 1921 1922 return 0; 1923 } 1924 1925 static struct sysctllog *vfsbio_sysctllog; 1926 1927 static void 1928 sysctl_kern_buf_setup(void) 1929 { 1930 1931 sysctl_createv(&vfsbio_sysctllog, 0, NULL, NULL, 1932 CTLFLAG_PERMANENT, 1933 CTLTYPE_NODE, "buf", 1934 SYSCTL_DESCR("Kernel buffer cache information"), 1935 sysctl_dobuf, 0, NULL, 0, 1936 CTL_KERN, KERN_BUF, CTL_EOL); 1937 } 1938 1939 static void 1940 sysctl_vm_buf_setup(void) 1941 { 1942 1943 sysctl_createv(&vfsbio_sysctllog, 0, NULL, NULL, 1944 CTLFLAG_PERMANENT|CTLFLAG_READWRITE, 1945 CTLTYPE_INT, "bufcache", 1946 SYSCTL_DESCR("Percentage of physical memory to use for " 1947 "buffer cache"), 1948 sysctl_bufvm_update, 0, &bufcache, 0, 1949 CTL_VM, CTL_CREATE, CTL_EOL); 1950 sysctl_createv(&vfsbio_sysctllog, 0, NULL, NULL, 1951 CTLFLAG_PERMANENT|CTLFLAG_READONLY, 1952 CTLTYPE_LONG, "bufmem", 1953 SYSCTL_DESCR("Amount of kernel memory used by buffer cache"), 1954 NULL, 0, &bufmem, 0, 1955 CTL_VM, CTL_CREATE, CTL_EOL); 1956 sysctl_createv(&vfsbio_sysctllog, 0, NULL, NULL, 1957 CTLFLAG_PERMANENT|CTLFLAG_READWRITE, 1958 CTLTYPE_LONG, "bufmem_lowater", 1959 SYSCTL_DESCR("Minimum amount of kernel memory to reserve for " 1960 "buffer cache"), 1961 sysctl_bufvm_update, 0, &bufmem_lowater, 0, 1962 CTL_VM, CTL_CREATE, CTL_EOL); 1963 sysctl_createv(&vfsbio_sysctllog, 0, NULL, NULL, 1964 CTLFLAG_PERMANENT|CTLFLAG_READWRITE, 1965 CTLTYPE_LONG, "bufmem_hiwater", 1966 SYSCTL_DESCR("Maximum amount of kernel memory to use for " 1967 "buffer cache"), 1968 sysctl_bufvm_update, 0, &bufmem_hiwater, 0, 1969 CTL_VM, CTL_CREATE, CTL_EOL); 1970 } 1971 1972 static int 1973 bufhash_stats(struct hashstat_sysctl *hs, bool fill) 1974 { 1975 buf_t *bp; 1976 uint64_t chain; 1977 1978 strlcpy(hs->hash_name, "bufhash", sizeof(hs->hash_name)); 1979 strlcpy(hs->hash_desc, "buffer hash", sizeof(hs->hash_desc)); 1980 if (!fill) 1981 return 0; 1982 1983 hs->hash_size = bufhash + 1; 1984 1985 for (size_t i = 0; i < hs->hash_size; i++) { 1986 chain = 0; 1987 1988 mutex_enter(&bufcache_lock); 1989 LIST_FOREACH(bp, &bufhashtbl[i], b_hash) { 1990 chain++; 1991 } 1992 mutex_exit(&bufcache_lock); 1993 1994 if (chain > 0) { 1995 hs->hash_used++; 1996 hs->hash_items += chain; 1997 if (chain > hs->hash_maxchain) 1998 hs->hash_maxchain = chain; 1999 } 2000 preempt_point(); 2001 } 2002 2003 return 0; 2004 } 2005 2006 #ifdef DEBUG 2007 /* 2008 * Print out statistics on the current allocation of the buffer pool. 2009 * Can be enabled to print out on every ``sync'' by setting "syncprt" 2010 * in vfs_syscalls.c using sysctl. 2011 */ 2012 void 2013 vfs_bufstats(void) 2014 { 2015 int i, j, count; 2016 buf_t *bp; 2017 struct bqueue *dp; 2018 int counts[MAXBSIZE / MIN_PAGE_SIZE + 1]; 2019 static const char *bname[BQUEUES] = { "LOCKED", "LRU", "AGE" }; 2020 2021 for (dp = bufqueues, i = 0; dp < &bufqueues[BQUEUES]; dp++, i++) { 2022 count = 0; 2023 memset(counts, 0, sizeof(counts)); 2024 TAILQ_FOREACH(bp, &dp->bq_queue, b_freelist) { 2025 counts[bp->b_bufsize / PAGE_SIZE]++; 2026 count++; 2027 } 2028 printf("%s: total-%d", bname[i], count); 2029 for (j = 0; j <= MAXBSIZE / PAGE_SIZE; j++) 2030 if (counts[j] != 0) 2031 printf(", %d-%d", j * PAGE_SIZE, counts[j]); 2032 printf("\n"); 2033 } 2034 } 2035 #endif /* DEBUG */ 2036 2037 /* ------------------------------ */ 2038 2039 buf_t * 2040 getiobuf(struct vnode *vp, bool waitok) 2041 { 2042 buf_t *bp; 2043 2044 bp = pool_cache_get(bufio_cache, (waitok ? PR_WAITOK : PR_NOWAIT)); 2045 if (bp == NULL) 2046 return bp; 2047 2048 buf_init(bp); 2049 2050 if ((bp->b_vp = vp) != NULL) { 2051 bp->b_objlock = vp->v_interlock; 2052 } else { 2053 KASSERT(bp->b_objlock == &buffer_lock); 2054 } 2055 2056 return bp; 2057 } 2058 2059 void 2060 putiobuf(buf_t *bp) 2061 { 2062 2063 buf_destroy(bp); 2064 pool_cache_put(bufio_cache, bp); 2065 } 2066 2067 /* 2068 * nestiobuf_iodone: b_iodone callback for nested buffers. 2069 */ 2070 2071 void 2072 nestiobuf_iodone(buf_t *bp) 2073 { 2074 buf_t *mbp = bp->b_private; 2075 int error; 2076 int donebytes; 2077 2078 KASSERT(bp->b_bcount <= bp->b_bufsize); 2079 KASSERT(mbp != bp); 2080 2081 error = bp->b_error; 2082 if (bp->b_error == 0 && 2083 (bp->b_bcount < bp->b_bufsize || bp->b_resid > 0)) { 2084 /* 2085 * Not all got transferred, raise an error. We have no way to 2086 * propagate these conditions to mbp. 2087 */ 2088 error = SET_ERROR(EIO); 2089 } 2090 2091 donebytes = bp->b_bufsize; 2092 2093 putiobuf(bp); 2094 nestiobuf_done(mbp, donebytes, error); 2095 } 2096 2097 /* 2098 * nestiobuf_setup: setup a "nested" buffer. 2099 * 2100 * => 'mbp' is a "master" buffer which is being divided into sub pieces. 2101 * => 'bp' should be a buffer allocated by getiobuf. 2102 * => 'offset' is a byte offset in the master buffer. 2103 * => 'size' is a size in bytes of this nested buffer. 2104 */ 2105 2106 void 2107 nestiobuf_setup(buf_t *mbp, buf_t *bp, int offset, size_t size) 2108 { 2109 const int b_pass = mbp->b_flags & (B_READ|B_PHYS|B_RAW|B_MEDIA_FLAGS); 2110 struct vnode *vp = mbp->b_vp; 2111 2112 KASSERT(mbp->b_bcount >= offset + size); 2113 bp->b_vp = vp; 2114 bp->b_dev = mbp->b_dev; 2115 bp->b_objlock = mbp->b_objlock; 2116 bp->b_cflags = BC_BUSY; 2117 bp->b_flags = B_ASYNC | b_pass; 2118 bp->b_iodone = nestiobuf_iodone; 2119 bp->b_data = (char *)mbp->b_data + offset; 2120 bp->b_resid = bp->b_bcount = size; 2121 bp->b_bufsize = bp->b_bcount; 2122 bp->b_private = mbp; 2123 BIO_COPYPRIO(bp, mbp); 2124 if (BUF_ISWRITE(bp) && vp != NULL) { 2125 mutex_enter(vp->v_interlock); 2126 vp->v_numoutput++; 2127 mutex_exit(vp->v_interlock); 2128 } 2129 } 2130 2131 /* 2132 * nestiobuf_done: propagate completion to the master buffer. 2133 * 2134 * => 'donebytes' specifies how many bytes in the 'mbp' is completed. 2135 * => 'error' is an errno(2) that 'donebytes' has been completed with. 2136 */ 2137 2138 void 2139 nestiobuf_done(buf_t *mbp, int donebytes, int error) 2140 { 2141 2142 if (donebytes == 0) { 2143 return; 2144 } 2145 mutex_enter(mbp->b_objlock); 2146 KASSERT(mbp->b_resid >= donebytes); 2147 mbp->b_resid -= donebytes; 2148 if (error) 2149 mbp->b_error = error; 2150 if (mbp->b_resid == 0) { 2151 if (mbp->b_error) 2152 mbp->b_resid = mbp->b_bcount; 2153 mutex_exit(mbp->b_objlock); 2154 biodone(mbp); 2155 } else 2156 mutex_exit(mbp->b_objlock); 2157 } 2158 2159 void 2160 buf_init(buf_t *bp) 2161 { 2162 2163 cv_init(&bp->b_busy, "biolock"); 2164 cv_init(&bp->b_done, "biowait"); 2165 bp->b_dev = NODEV; 2166 bp->b_error = 0; 2167 bp->b_flags = 0; 2168 bp->b_cflags = 0; 2169 bp->b_oflags = 0; 2170 bp->b_objlock = &buffer_lock; 2171 bp->b_iodone = NULL; 2172 bp->b_dev = NODEV; 2173 bp->b_vnbufs.le_next = NOLIST; 2174 BIO_SETPRIO(bp, BPRIO_DEFAULT); 2175 } 2176 2177 void 2178 buf_destroy(buf_t *bp) 2179 { 2180 2181 cv_destroy(&bp->b_done); 2182 cv_destroy(&bp->b_busy); 2183 } 2184 2185 int 2186 bbusy(buf_t *bp, bool intr, int timo, kmutex_t *interlock) 2187 { 2188 int error; 2189 2190 KASSERT(mutex_owned(&bufcache_lock)); 2191 2192 SDT_PROBE4(io, kernel, , bbusy__start, bp, intr, timo, interlock); 2193 2194 if ((bp->b_cflags & BC_BUSY) != 0) { 2195 if (curlwp == uvm.pagedaemon_lwp) { 2196 error = SET_ERROR(EDEADLK); 2197 goto out; 2198 } 2199 bp->b_cflags |= BC_WANTED; 2200 if (interlock != NULL) 2201 mutex_exit(interlock); 2202 if (intr) { 2203 error = cv_timedwait_sig(&bp->b_busy, &bufcache_lock, 2204 timo); 2205 } else { 2206 error = cv_timedwait(&bp->b_busy, &bufcache_lock, 2207 timo); 2208 } 2209 /* 2210 * At this point the buffer may be gone: don't touch it 2211 * again. The caller needs to find it again and retry. 2212 */ 2213 if (interlock != NULL) 2214 mutex_enter(interlock); 2215 if (error == 0) 2216 error = SET_ERROR(EPASSTHROUGH); 2217 } else { 2218 bp->b_cflags |= BC_BUSY; 2219 error = 0; 2220 } 2221 2222 out: SDT_PROBE5(io, kernel, , bbusy__done, 2223 bp, intr, timo, interlock, error); 2224 return error; 2225 } 2226 2227 /* 2228 * Nothing outside this file should really need to know about nbuf, 2229 * but a few things still want to read it, so give them a way to do that. 2230 */ 2231 u_int 2232 buf_nbuf(void) 2233 { 2234 2235 return nbuf; 2236 } 2237
Indexes created Sat Sep 20 22:09:52 GMT 2025