xref: /freebsd/stand/common/bcache.c (revision 6137b5f7b8c183ee8806d79b3f1d8e5e3ddb3df3)
1 /*-
2  * Copyright (c) 1998 Michael Smith <msmith@freebsd.org>
3  * Copyright 2015 Toomas Soome <tsoome@me.com>
4  * All rights reserved.
5  *
6  * Redistribution and use in source and binary forms, with or without
7  * modification, are permitted provided that the following conditions
8  * are met:
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.
14  *
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
25  * SUCH DAMAGE.
26  */
27 
28 #include <sys/param.h>
29 /*
30  * Simple hashed block cache
31  */
32 
33 #include <sys/stdint.h>
34 
35 #include <stand.h>
36 #include <string.h>
37 #include <strings.h>
38 
39 #include "bootstrap.h"
40 
41 /* #define BCACHE_DEBUG */
42 
43 #ifdef BCACHE_DEBUG
44 # define DPRINTF(fmt, args...)	printf("%s: " fmt "\n" , __func__ , ## args)
45 #else
46 # define DPRINTF(fmt, args...)	((void)0)
47 #endif
48 
49 struct bcachectl
50 {
51     daddr_t	bc_blkno;
52     int		bc_count;
53 };
54 
55 /*
56  * bcache per device node. cache is allocated on device first open and freed
57  * on last close, to save memory. The issue there is the size; biosdisk
58  * supports up to 31 (0x1f) devices. Classic setup would use single disk
59  * to boot from, but this has changed with zfs.
60  */
61 struct bcache {
62     struct bcachectl	*bcache_ctl;
63     caddr_t		bcache_data;
64     size_t		bcache_nblks;
65     size_t		ra;
66     daddr_t		bcache_nextblkno;
67     size_t		ralen;
68 };
69 
70 static u_int bcache_total_nblks;	/* set by bcache_init */
71 static u_int bcache_blksize;		/* set by bcache_init */
72 static u_int bcache_numdev;		/* set by bcache_add_dev */
73 /* statistics */
74 static u_int bcache_units;	/* number of devices with cache */
75 static u_int bcache_unit_nblks;	/* nblocks per unit */
76 static u_int bcache_hits;
77 static u_int bcache_misses;
78 static u_int bcache_ops;
79 static u_int bcache_bypasses;
80 static u_int bcache_bcount;
81 static u_int bcache_rablks;
82 
83 #define	BHASH(bc, blkno)	((blkno) & ((bc)->bcache_nblks - 1))
84 #define	BCACHE_LOOKUP(bc, blkno)	\
85 	((bc)->bcache_ctl[BHASH((bc), (blkno))].bc_blkno != (blkno))
86 #define	BCACHE_READAHEAD	512
87 #define	BCACHE_MINREADAHEAD	32
88 #define	BCACHE_MAXIOWRA		512
89 
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);
93 
94 /*
95  * Initialise the cache for (nblks) of (bsize).
96  */
97 void
98 bcache_init(size_t nblks, size_t bsize)
99 {
100     /* set up control data */
101     bcache_total_nblks = nblks;
102     bcache_blksize = bsize;
103 }
104 
105 /*
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
111  * for zfs, last.
112  */
113 void
114 bcache_add_dev(int devices)
115 {
116     bcache_numdev += devices;
117 }
118 
119 void *
120 bcache_allocate(void)
121 {
122     u_int i;
123     struct bcache *bc = malloc(sizeof (struct bcache));
124     int disks = bcache_numdev;
125 
126     if (disks == 0)
127 	disks = 1;	/* safe guard */
128 
129     if (bc == NULL) {
130 	errno = ENOMEM;
131 	return (bc);
132     }
133 
134     /*
135      * the bcache block count must be power of 2 for hash function
136      */
137     i = fls(disks) - 1;		/* highbit - 1 */
138     if (disks > (1 << i))	/* next power of 2 */
139 	i++;
140 
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 +
148 	sizeof(uint32_t));
149     }
150 
151     bc->bcache_ctl = malloc(bc->bcache_nblks * sizeof(struct bcachectl));
152 
153     if ((bc->bcache_data == NULL) || (bc->bcache_ctl == NULL)) {
154 	bcache_free_instance(bc);
155 	errno = ENOMEM;
156 	return (NULL);
157     }
158 
159     /* Flush the cache */
160     for (i = 0; i < bc->bcache_nblks; i++) {
161 	bc->bcache_ctl[i].bc_count = -1;
162 	bc->bcache_ctl[i].bc_blkno = -1;
163     }
164     bcache_units++;
165     bc->ra = BCACHE_READAHEAD;	/* optimistic read ahead */
166     bc->bcache_nextblkno = -1;
167     return (bc);
168 }
169 
170 void
171 bcache_free(void *cache)
172 {
173     struct bcache *bc = cache;
174 
175     if (bc == NULL)
176 	return;
177 
178     bcache_free_instance(bc);
179     bcache_units--;
180 }
181 
182 /*
183  * Handle a write request; write directly to the disk, and populate the
184  * cache with the new values.
185  */
186 static int
187 write_strategy(void *devdata, int rw, daddr_t blk, size_t size,
188     char *buf, size_t *rsize)
189 {
190     struct bcache_devdata	*dd = (struct bcache_devdata *)devdata;
191     struct bcache		*bc = dd->dv_cache;
192     daddr_t			i, nblk;
193 
194     nblk = size / bcache_blksize;
195 
196     /* Invalidate the blocks being written */
197     for (i = 0; i < nblk; i++) {
198 	bcache_invalidate(bc, blk + i);
199     }
200 
201     /* Write the blocks */
202     return (dd->dv_strategy(dd->dv_devdata, rw, blk, size, buf, rsize));
203 }
204 
205 /*
206  * Handle a read request; fill in parts of the request that can
207  * be satisfied by the cache, use the supplied strategy routine to do
208  * device I/O and then use the I/O results to populate the cache.
209  */
210 static int
211 read_strategy(void *devdata, int rw, daddr_t blk, size_t size,
212     char *buf, size_t *rsize)
213 {
214     struct bcache_devdata	*dd = (struct bcache_devdata *)devdata;
215     struct bcache		*bc = dd->dv_cache;
216     size_t			i, nblk, p_size, r_size, complete, ra;
217     int				result;
218     daddr_t			p_blk;
219     caddr_t			p_buf;
220 
221     if (bc == NULL) {
222 	errno = ENODEV;
223 	return (-1);
224     }
225 
226     if (rsize != NULL)
227 	*rsize = 0;
228 
229     nblk = size / bcache_blksize;
230     if (nblk == 0 && size != 0)
231 	nblk++;
232     result = 0;
233     complete = 1;
234 
235     /* Satisfy any cache hits up front, break on first miss */
236     for (i = 0; i < nblk; i++) {
237 	if (BCACHE_LOOKUP(bc, (daddr_t)(blk + i))) {
238 	    bcache_misses += (nblk - i);
239 	    complete = 0;
240 	    break;
241 	} else {
242 	    bcache_hits++;
243 	}
244     }
245 
246     /*
247      * Adjust read-ahead size if appropriate.  Subject to the requirement
248      * that bc->ra must stay in between MINREADAHEAD and READAHEAD, we
249      * increase it when we notice that readahead was useful and decrease
250      * it when we notice that readahead was not useful.
251      */
252     if (complete || (i == bc->ralen && bc->ralen > 0)) {
253 	if (bc->ra < BCACHE_READAHEAD)
254 	    bc->ra <<= 1;	/* increase read ahead */
255     } else {
256 	if (nblk - i > BCACHE_MINREADAHEAD && bc->ralen > 0 &&
257 	  bc->ra > BCACHE_MINREADAHEAD)
258 	    bc->ra >>= 1;	/* reduce read ahead */
259     }
260 
261     /* Adjust our "unconsumed readahead" value. */
262     if (blk == bc->bcache_nextblkno) {
263 	if (nblk > bc->ralen)
264 	    bc->ralen = 0;
265 	else
266 	    bc->ralen -= nblk;
267     }
268 
269     if (complete) {	/* whole set was in cache, return it */
270 	bcopy(bc->bcache_data + (bcache_blksize * BHASH(bc, blk)), buf, size);
271 	goto done;
272     }
273 
274     /*
275      * Fill in any misses. From check we have i pointing to first missing
276      * block, read in all remaining blocks + readahead.
277      * We have space at least for nblk - i before bcache wraps.
278      */
279     p_blk = blk + i;
280     p_buf = bc->bcache_data + (bcache_blksize * BHASH(bc, p_blk));
281     r_size = bc->bcache_nblks - BHASH(bc, p_blk); /* remaining blocks */
282 
283     p_size = MIN(r_size, nblk - i);	/* read at least those blocks */
284 
285     /*
286      * The read ahead size setup.
287      * While the read ahead can save us IO, it also can complicate things:
288      * 1. We do not want to read ahead by wrapping around the
289      * bcache end - this would complicate the cache management.
290      * 2. We are using bc->ra as dynamic hint for read ahead size,
291      * detected cache hits will increase the read-ahead block count, and
292      * misses will decrease, see the code above.
293      * 3. The bcache is sized by 512B blocks, however, the underlying device
294      * may have a larger sector size, and we should perform the IO by
295      * taking into account these larger sector sizes. We could solve this by
296      * passing the sector size to bcache_allocate(), or by using ioctl(), but
297      * in this version we are using the constant, 16 blocks, and are rounding
298      * read ahead block count down to multiple of 16.
299      * Using the constant has two reasons, we are not entirely sure if the
300      * BIOS disk interface is providing the correct value for sector size.
301      * And secondly, this way we get the most conservative setup for the ra.
302      *
303      * The selection of multiple of 16 blocks (8KB) is quite arbitrary, however,
304      * we want to cover CDs (2K) and 4K disks.
305      * bcache_allocate() will always fall back to a minimum of 32 blocks.
306      * Our choice of 16 read ahead blocks will always fit inside the bcache.
307      */
308 
309     if ((rw & F_NORA) == F_NORA)
310 	ra = 0;
311     else
312 	ra = bc->bcache_nblks - BHASH(bc, p_blk + p_size);
313 
314     /*
315      * Only trigger read-ahead if we detect two blocks being read
316      * sequentially.
317      */
318     if ((bc->bcache_nextblkno != blk) && ra != 0) {
319         ra = 0;
320     }
321 
322     if (ra != 0 && ra != bc->bcache_nblks) { /* do we have RA space? */
323 	ra = MIN(bc->ra, ra - 1);
324 	ra = rounddown(ra, 16);		/* multiple of 16 blocks */
325 	if (ra + p_size > BCACHE_MAXIOWRA)
326 	    ra = BCACHE_MAXIOWRA - p_size;
327 	bc->ralen = ra;
328 	p_size += ra;
329     } else {
330 	bc->ralen = 0;
331     }
332 
333     /* invalidate bcache */
334     for (i = 0; i < p_size; i++) {
335 	bcache_invalidate(bc, p_blk + i);
336     }
337 
338     r_size = 0;
339     /*
340      * with read-ahead, it may happen we are attempting to read past
341      * disk end, as bcache has no information about disk size.
342      * in such case we should get partial read if some blocks can be
343      * read or error, if no blocks can be read.
344      * in either case we should return the data in bcache and only
345      * return error if there is no data.
346      */
347     rw &= F_MASK;
348     result = dd->dv_strategy(dd->dv_devdata, rw, p_blk,
349 	p_size * bcache_blksize, p_buf, &r_size);
350 
351     r_size /= bcache_blksize;
352     for (i = 0; i < r_size; i++)
353 	bcache_insert(bc, p_blk + i);
354 
355     /* update ra statistics */
356     if (r_size != 0) {
357 	if (r_size < p_size)
358 	    bcache_rablks += (p_size - r_size);
359 	else
360 	    bcache_rablks += ra;
361     }
362 
363     /* check how much data can we copy */
364     for (i = 0; i < nblk; i++) {
365 	if (BCACHE_LOOKUP(bc, (daddr_t)(blk + i)))
366 	    break;
367     }
368 
369     if (size > i * bcache_blksize)
370 	size = i * bcache_blksize;
371 
372     if (size != 0) {
373 	bcopy(bc->bcache_data + (bcache_blksize * BHASH(bc, blk)), buf, size);
374 	result = 0;
375     }
376 
377 done:
378     if (result == 0) {
379         if (rsize != NULL)
380 	    *rsize = size;
381         bc->bcache_nextblkno = blk + (size / DEV_BSIZE);
382     }
383     return(result);
384 }
385 
386 /*
387  * Requests larger than 1/2 cache size will be bypassed and go
388  * directly to the disk.  XXX tune this.
389  */
390 int
391 bcache_strategy(void *devdata, int rw, daddr_t blk, size_t size,
392     char *buf, size_t *rsize)
393 {
394     struct bcache_devdata	*dd = (struct bcache_devdata *)devdata;
395     struct bcache		*bc = dd->dv_cache;
396     u_int bcache_nblks = 0;
397     int nblk, cblk, ret;
398     size_t csize, isize, total;
399 
400     bcache_ops++;
401 
402     if (bc != NULL)
403 	bcache_nblks = bc->bcache_nblks;
404 
405     /* bypass large requests, or when the cache is inactive */
406     if (bc == NULL ||
407 	((size * 2 / bcache_blksize) > bcache_nblks)) {
408 	DPRINTF("bypass %zu from %jd", size / bcache_blksize, blk);
409 	bcache_bypasses++;
410 	rw &= F_MASK;
411 	return (dd->dv_strategy(dd->dv_devdata, rw, blk, size, buf, rsize));
412     }
413 
414     switch (rw & F_MASK) {
415     case F_READ:
416 	nblk = size / bcache_blksize;
417 	if (size != 0 && nblk == 0)
418 	    nblk++;	/* read at least one block */
419 
420 	ret = 0;
421 	total = 0;
422 	while(size) {
423 	    cblk = bcache_nblks - BHASH(bc, blk); /* # of blocks left */
424 	    cblk = MIN(cblk, nblk);
425 
426 	    if (size <= bcache_blksize)
427 		csize = size;
428 	    else
429 		csize = cblk * bcache_blksize;
430 
431 	    ret = read_strategy(devdata, rw, blk, csize, buf+total, &isize);
432 
433 	    /*
434 	     * we may have error from read ahead, if we have read some data
435 	     * return partial read.
436 	     */
437 	    if (ret != 0 || isize == 0) {
438 		if (total != 0)
439 		    ret = 0;
440 		break;
441 	    }
442 	    blk += isize / bcache_blksize;
443 	    total += isize;
444 	    size -= isize;
445 	    nblk = size / bcache_blksize;
446 	}
447 
448 	if (rsize)
449 	    *rsize = total;
450 
451 	return (ret);
452     case F_WRITE:
453 	return write_strategy(devdata, F_WRITE, blk, size, buf, rsize);
454     }
455     return -1;
456 }
457 
458 /*
459  * Free allocated bcache instance
460  */
461 static void
462 bcache_free_instance(struct bcache *bc)
463 {
464     if (bc != NULL) {
465 	free(bc->bcache_ctl);
466 	free(bc->bcache_data);
467 	free(bc);
468     }
469 }
470 
471 /*
472  * Insert a block into the cache.
473  */
474 static void
475 bcache_insert(struct bcache *bc, daddr_t blkno)
476 {
477     u_int	cand;
478 
479     cand = BHASH(bc, blkno);
480 
481     DPRINTF("insert blk %jd -> %u # %d", blkno, cand, bcache_bcount);
482     bc->bcache_ctl[cand].bc_blkno = blkno;
483     bc->bcache_ctl[cand].bc_count = bcache_bcount++;
484 }
485 
486 /*
487  * Invalidate a block from the cache.
488  */
489 static void
490 bcache_invalidate(struct bcache *bc, daddr_t blkno)
491 {
492     u_int	i;
493 
494     i = BHASH(bc, blkno);
495     if (bc->bcache_ctl[i].bc_blkno == blkno) {
496 	bc->bcache_ctl[i].bc_count = -1;
497 	bc->bcache_ctl[i].bc_blkno = -1;
498 	DPRINTF("invalidate blk %ju", blkno);
499     }
500 }
501 
502 #ifndef BOOT2
503 COMMAND_SET(bcachestat, "bcachestat", "get disk block cache stats", command_bcache);
504 
505 static int
506 command_bcache(int argc, char *argv[] __unused)
507 {
508     if (argc != 1) {
509 	command_errmsg = "wrong number of arguments";
510 	return(CMD_ERROR);
511     }
512 
513     printf("\ncache blocks: %u\n", bcache_total_nblks);
514     printf("cache blocksz: %u\n", bcache_blksize);
515     printf("cache readahead: %u\n", bcache_rablks);
516     printf("unit cache blocks: %u\n", bcache_unit_nblks);
517     printf("cached units: %u\n", bcache_units);
518     printf("%u ops %d bypasses %u hits %u misses\n", bcache_ops,
519 	bcache_bypasses, bcache_hits, bcache_misses);
520     return(CMD_OK);
521 }
522 #endif
523