2 * Copyright (c) 1998 Michael Smith <msmith@freebsd.org>
3 * Copyright 2015 Toomas Soome <tsoome@me.com>
6 * Redistribution and use in source and binary forms, with or without
7 * modification, are permitted provided that the following conditions
9 * 1. Redistributions of source code must retain the above copyright
10 * notice, this list of conditions and the following disclaimer.
11 * 2. Redistributions in binary form must reproduce the above copyright
12 * notice, this list of conditions and the following disclaimer in the
13 * documentation and/or other materials provided with the distribution.
15 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
16 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
17 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
18 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
19 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
20 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
21 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
22 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
23 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
24 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
28 #include <sys/cdefs.h>
29 #include <sys/param.h>
30 __FBSDID("$FreeBSD$");
33 * Simple hashed block cache
36 #include <sys/stdint.h>
42 #include "bootstrap.h"
44 /* #define BCACHE_DEBUG */
47 # define DEBUG(fmt, args...) printf("%s: " fmt "\n" , __func__ , ## args)
49 # define DEBUG(fmt, args...)
59 * bcache per device node. cache is allocated on device first open and freed
60 * on last close, to save memory. The issue there is the size; biosdisk
61 * supports up to 31 (0x1f) devices. Classic setup would use single disk
62 * to boot from, but this has changed with zfs.
65 struct bcachectl *bcache_ctl;
71 static u_int bcache_total_nblks; /* set by bcache_init */
72 static u_int bcache_blksize; /* set by bcache_init */
73 static u_int bcache_numdev; /* set by bcache_add_dev */
75 static u_int bcache_units; /* number of devices with cache */
76 static u_int bcache_unit_nblks; /* nblocks per unit */
77 static u_int bcache_hits;
78 static u_int bcache_misses;
79 static u_int bcache_ops;
80 static u_int bcache_bypasses;
81 static u_int bcache_bcount;
82 static u_int bcache_rablks;
84 #define BHASH(bc, blkno) ((blkno) & ((bc)->bcache_nblks - 1))
85 #define BCACHE_LOOKUP(bc, blkno) \
86 ((bc)->bcache_ctl[BHASH((bc), (blkno))].bc_blkno != (blkno))
87 #define BCACHE_READAHEAD 256
88 #define BCACHE_MINREADAHEAD 32
90 static void bcache_invalidate(struct bcache *bc, daddr_t blkno);
91 static void bcache_insert(struct bcache *bc, daddr_t blkno);
92 static void bcache_free_instance(struct bcache *bc);
95 * Initialise the cache for (nblks) of (bsize).
98 bcache_init(u_int nblks, size_t bsize)
100 /* set up control data */
101 bcache_total_nblks = nblks;
102 bcache_blksize = bsize;
106 * add number of devices to bcache. we have to divide cache space
107 * between the devices, so bcache_add_dev() can be used to set up the
108 * number. The issue is, we need to get the number before actual allocations.
109 * bcache_add_dev() is supposed to be called from device init() call, so the
110 * assumption is, devsw dv_init is called for plain devices first, and
114 bcache_add_dev(int devices)
116 bcache_numdev += devices;
120 bcache_allocate(void)
123 struct bcache *bc = malloc(sizeof (struct bcache));
124 int disks = bcache_numdev;
127 disks = 1; /* safe guard */
135 * the bcache block count must be power of 2 for hash function
137 i = fls(disks) - 1; /* highbit - 1 */
138 if (disks > (1 << i)) /* next power of 2 */
141 bc->bcache_nblks = bcache_total_nblks >> i;
142 bcache_unit_nblks = bc->bcache_nblks;
143 bc->bcache_data = malloc(bc->bcache_nblks * bcache_blksize);
144 if (bc->bcache_data == NULL) {
145 /* dont error out yet. fall back to 32 blocks and try again */
146 bc->bcache_nblks = 32;
147 bc->bcache_data = malloc(bc->bcache_nblks * bcache_blksize);
150 bc->bcache_ctl = malloc(bc->bcache_nblks * sizeof(struct bcachectl));
152 if ((bc->bcache_data == NULL) || (bc->bcache_ctl == NULL)) {
153 bcache_free_instance(bc);
158 /* Flush the cache */
159 for (i = 0; i < bc->bcache_nblks; i++) {
160 bc->bcache_ctl[i].bc_count = -1;
161 bc->bcache_ctl[i].bc_blkno = -1;
164 bc->ra = BCACHE_READAHEAD; /* optimistic read ahead */
169 bcache_free(void *cache)
171 struct bcache *bc = cache;
176 bcache_free_instance(bc);
181 * Handle a write request; write directly to the disk, and populate the
182 * cache with the new values.
185 write_strategy(void *devdata, int rw, daddr_t blk, size_t offset,
186 size_t size, char *buf, size_t *rsize)
188 struct bcache_devdata *dd = (struct bcache_devdata *)devdata;
189 struct bcache *bc = dd->dv_cache;
192 nblk = size / bcache_blksize;
194 /* Invalidate the blocks being written */
195 for (i = 0; i < nblk; i++) {
196 bcache_invalidate(bc, blk + i);
199 /* Write the blocks */
200 return (dd->dv_strategy(dd->dv_devdata, rw, blk, offset, size, buf, rsize));
204 * Handle a read request; fill in parts of the request that can
205 * be satisfied by the cache, use the supplied strategy routine to do
206 * device I/O and then use the I/O results to populate the cache.
209 read_strategy(void *devdata, int rw, daddr_t blk, size_t offset,
210 size_t size, char *buf, size_t *rsize)
212 struct bcache_devdata *dd = (struct bcache_devdata *)devdata;
213 struct bcache *bc = dd->dv_cache;
214 size_t i, nblk, p_size, r_size, complete, ra;
227 nblk = size / bcache_blksize;
228 if ((nblk == 0 && size != 0) || offset != 0)
233 /* Satisfy any cache hits up front, break on first miss */
234 for (i = 0; i < nblk; i++) {
235 if (BCACHE_LOOKUP(bc, (daddr_t)(blk + i))) {
236 bcache_misses += (nblk - i);
238 if (nblk - i > BCACHE_MINREADAHEAD && bc->ra > BCACHE_MINREADAHEAD)
239 bc->ra >>= 1; /* reduce read ahead */
246 if (complete) { /* whole set was in cache, return it */
247 if (bc->ra < BCACHE_READAHEAD)
248 bc->ra <<= 1; /* increase read ahead */
249 bcopy(bc->bcache_data + (bcache_blksize * BHASH(bc, blk)) + offset,
255 * Fill in any misses. From check we have i pointing to first missing
256 * block, read in all remaining blocks + readahead.
257 * We have space at least for nblk - i before bcache wraps.
260 p_buf = bc->bcache_data + (bcache_blksize * BHASH(bc, p_blk));
261 r_size = bc->bcache_nblks - BHASH(bc, p_blk); /* remaining blocks */
263 p_size = MIN(r_size, nblk - i); /* read at least those blocks */
265 ra = bc->bcache_nblks - BHASH(bc, p_blk + p_size);
266 if (ra != bc->bcache_nblks) { /* do we have RA space? */
267 ra = MIN(bc->ra, ra);
271 /* invalidate bcache */
272 for (i = 0; i < p_size; i++) {
273 bcache_invalidate(bc, p_blk + i);
278 * with read-ahead, it may happen we are attempting to read past
279 * disk end, as bcache has no information about disk size.
280 * in such case we should get partial read if some blocks can be
281 * read or error, if no blocks can be read.
282 * in either case we should return the data in bcache and only
283 * return error if there is no data.
285 result = dd->dv_strategy(dd->dv_devdata, rw, p_blk, 0,
286 p_size * bcache_blksize, p_buf, &r_size);
288 r_size /= bcache_blksize;
289 for (i = 0; i < r_size; i++)
290 bcache_insert(bc, p_blk + i);
292 /* update ra statistics */
295 bcache_rablks += (p_size - r_size);
300 /* check how much data can we copy */
301 for (i = 0; i < nblk; i++) {
302 if (BCACHE_LOOKUP(bc, (daddr_t)(blk + i)))
306 if (size > i * bcache_blksize)
307 size = i * bcache_blksize;
310 bcopy(bc->bcache_data + (bcache_blksize * BHASH(bc, blk)) + offset,
316 if ((result == 0) && (rsize != NULL))
322 * Requests larger than 1/2 cache size will be bypassed and go
323 * directly to the disk. XXX tune this.
326 bcache_strategy(void *devdata, int rw, daddr_t blk, size_t offset,
327 size_t size, char *buf, size_t *rsize)
329 struct bcache_devdata *dd = (struct bcache_devdata *)devdata;
330 struct bcache *bc = dd->dv_cache;
331 u_int bcache_nblks = 0;
333 size_t csize, isize, total;
338 bcache_nblks = bc->bcache_nblks;
340 /* bypass large requests, or when the cache is inactive */
342 (offset == 0 && ((size * 2 / bcache_blksize) > bcache_nblks))) {
343 DEBUG("bypass %d from %d", size / bcache_blksize, blk);
345 return (dd->dv_strategy(dd->dv_devdata, rw, blk, offset, size, buf,
349 /* normalize offset */
350 while (offset >= bcache_blksize) {
352 offset -= bcache_blksize;
357 nblk = size / bcache_blksize;
358 if (offset || (size != 0 && nblk == 0))
359 nblk++; /* read at least one block */
364 cblk = bcache_nblks - BHASH(bc, blk); /* # of blocks left */
365 cblk = MIN(cblk, nblk);
367 if (size <= bcache_blksize)
370 csize = cblk * bcache_blksize;
372 csize -= (bcache_blksize - offset);
375 ret = read_strategy(devdata, rw, blk, offset,
376 csize, buf+total, &isize);
379 * we may have error from read ahead, if we have read some data
380 * return partial read.
382 if (ret != 0 || isize == 0) {
387 blk += (offset+isize) / bcache_blksize;
391 nblk = size / bcache_blksize;
399 return write_strategy(devdata, rw, blk, offset, size, buf, rsize);
405 * Free allocated bcache instance
408 bcache_free_instance(struct bcache *bc)
412 free(bc->bcache_ctl);
414 free(bc->bcache_data);
420 * Insert a block into the cache.
423 bcache_insert(struct bcache *bc, daddr_t blkno)
427 cand = BHASH(bc, blkno);
429 DEBUG("insert blk %llu -> %u # %d", blkno, cand, bcache_bcount);
430 bc->bcache_ctl[cand].bc_blkno = blkno;
431 bc->bcache_ctl[cand].bc_count = bcache_bcount++;
435 * Invalidate a block from the cache.
438 bcache_invalidate(struct bcache *bc, daddr_t blkno)
442 i = BHASH(bc, blkno);
443 if (bc->bcache_ctl[i].bc_blkno == blkno) {
444 bc->bcache_ctl[i].bc_count = -1;
445 bc->bcache_ctl[i].bc_blkno = -1;
446 DEBUG("invalidate blk %llu", blkno);
451 COMMAND_SET(bcachestat, "bcachestat", "get disk block cache stats", command_bcache);
454 command_bcache(int argc, char *argv[])
457 command_errmsg = "wrong number of arguments";
461 printf("\ncache blocks: %d\n", bcache_total_nblks);
462 printf("cache blocksz: %d\n", bcache_blksize);
463 printf("cache readahead: %d\n", bcache_rablks);
464 printf("unit cache blocks: %d\n", bcache_unit_nblks);
465 printf("cached units: %d\n", bcache_units);
466 printf("%d ops %d bypasses %d hits %d misses\n", bcache_ops,
467 bcache_bypasses, bcache_hits, bcache_misses);