xref: /titanic_41/usr/src/uts/common/sys/buf.h (revision b3697b90e692e3e5d859fb77d285d4c056d99eda)
1 /*
2  * CDDL HEADER START
3  *
4  * The contents of this file are subject to the terms of the
5  * Common Development and Distribution License (the "License").
6  * You may not use this file except in compliance with the License.
7  *
8  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9  * or http://www.opensolaris.org/os/licensing.
10  * See the License for the specific language governing permissions
11  * and limitations under the License.
12  *
13  * When distributing Covered Code, include this CDDL HEADER in each
14  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15  * If applicable, add the following below this CDDL HEADER, with the
16  * fields enclosed by brackets "[]" replaced with your own identifying
17  * information: Portions Copyright [yyyy] [name of copyright owner]
18  *
19  * CDDL HEADER END
20  */
21 /*
22  * Copyright 2008 Sun Microsystems, Inc.  All rights reserved.
23  * Use is subject to license terms.
24  */
25 
26 /*	Copyright (c) 1984, 1986, 1987, 1988, 1989 AT&T	*/
27 /*	  All Rights Reserved  	*/
28 
29 /*
30  * University Copyright- Copyright (c) 1982, 1986, 1988
31  * The Regents of the University of California
32  * All Rights Reserved
33  *
34  * University Acknowledgment- Portions of this document are derived from
35  * software developed by the University of California, Berkeley, and its
36  * contributors.
37  */
38 
39 #ifndef _SYS_BUF_H
40 #define	_SYS_BUF_H
41 
42 #include <sys/types32.h>
43 #include <sys/t_lock.h>
44 #include <sys/kstat.h>
45 
46 #ifdef	__cplusplus
47 extern "C" {
48 #endif
49 
50 /*
51  *	Each buffer in the pool is usually doubly linked into 2 lists:
52  *	the device with which it is currently associated (always)
53  *	and also on a list of blocks available for allocation
54  *	for other use (usually).
55  *	The latter list is kept in last-used order, and the two
56  *	lists are doubly linked to make it easy to remove
57  *	a buffer from one list when it was found by
58  *	looking through the other.
59  *	A buffer is on the available list, and is liable
60  *	to be reassigned to another disk block, if and only
61  *	if it is not marked BUSY.  When a buffer is busy, the
62  *	available-list pointers can be used for other purposes.
63  *	Most drivers use the forward ptr as a link in their I/O active queue.
64  *	A buffer header contains all the information required to perform I/O.
65  *	Most of the routines which manipulate these things are in bio.c.
66  *
67  *	There are a number of locks associated with the buffer management
68  *	system.
69  *	hbuf.b_lock:	protects hash chains, buffer hdr freelists
70  *			and delayed write freelist
71  *	bfree_lock;	protects the bfreelist structure
72  *	bhdr_lock:	protects the free header list
73  *	blist_lock:	protects b_list fields
74  *	buf.b_sem:	protects all remaining members in the buf struct
75  *	buf.b_io:	I/O synchronization variable
76  *
77  *	A buffer header is never "locked" (b_sem) when it is on
78  *	a "freelist" (bhdrlist or bfreelist avail lists).
79  */
80 typedef struct	buf {
81 	int	b_flags;		/* see defines below */
82 	struct buf *b_forw;		/* headed by d_tab of conf.c */
83 	struct buf *b_back;		/*  "  */
84 	struct buf *av_forw;		/* position on free list, */
85 	struct buf *av_back;		/* if not BUSY */
86 	o_dev_t	b_dev;			/* OLD major+minor device name */
87 	size_t b_bcount;		/* transfer count */
88 	union {
89 		caddr_t b_addr;		/* low order core address */
90 		struct fs *b_fs;	/* superblocks */
91 		struct cg *b_cg;	/* UFS cylinder group block */
92 		struct dinode *b_dino;	/* UFS ilist */
93 		daddr32_t *b_daddr;	/* disk blocks */
94 	} b_un;
95 
96 	lldaddr_t	_b_blkno;	/* block # on device (union) */
97 #define	b_lblkno	_b_blkno._f
98 #ifdef _LP64
99 #define	b_blkno		_b_blkno._f
100 #else
101 #define	b_blkno		_b_blkno._p._l
102 #endif /* _LP64 */
103 
104 	char	b_obs1;			/* obsolete */
105 	size_t	b_resid;		/* words not transferred after error */
106 	clock_t	b_start;		/* request start time */
107 	struct  proc  *b_proc;		/* process doing physical or swap I/O */
108 	struct	page  *b_pages;		/* page list for PAGEIO */
109 	clock_t b_obs2;			/* obsolete */
110 	/* Begin new stuff */
111 #define	b_actf	av_forw
112 #define	b_actl	av_back
113 #define	b_active b_bcount
114 #define	b_errcnt b_resid
115 	size_t	b_bufsize;		/* size of allocated buffer */
116 	int	(*b_iodone)(struct buf *);	/* function called by iodone */
117 	struct	vnode *b_vp;		/* vnode associated with block */
118 	struct 	buf *b_chain;		/* chain together all buffers here */
119 	int	b_obs3;			/* obsolete */
120 	int	b_error;		/* expanded error field */
121 	void	*b_private;		/* "opaque" driver private area */
122 	dev_t	b_edev;			/* expanded dev field */
123 	ksema_t	b_sem;			/* Exclusive access to buf */
124 	ksema_t	b_io;			/* I/O Synchronization */
125 	struct buf *b_list;		/* List of potential B_DELWRI bufs */
126 	struct page **b_shadow;		/* shadow page list */
127 	void	*b_dip;			/* device info pointer */
128 	struct vnode *b_file;		/* file associated with this buffer */
129 	offset_t b_offset;		/* offset in file assoc. with buffer */
130 } buf_t;
131 
132 /*
133  * Bufhd structures used at the head of the hashed buffer queues.
134  * We only need seven words for this, so this abbreviated
135  * definition saves some space.
136  */
137 struct diskhd {
138 	int	b_flags;		/* not used, needed for consistency */
139 	struct buf *b_forw, *b_back;	/* queue of unit queues */
140 	struct buf *av_forw, *av_back;	/* queue of bufs for this unit */
141 	o_dev_t	b_dev;			/* OLD major+minor device name */
142 	size_t b_bcount;		/* transfer count */
143 };
144 
145 
146 /*
147  * Statistics on the buffer cache
148  */
149 struct biostats {
150 	kstat_named_t	bio_lookup;	/* requests to assign buffer */
151 	kstat_named_t	bio_hit;	/* buffer already associated with blk */
152 	kstat_named_t	bio_bufwant;	/* kmem_allocs NOSLEEP failed new buf */
153 	kstat_named_t	bio_bufwait;	/* kmem_allocs with KM_SLEEP for buf */
154 	kstat_named_t	bio_bufbusy;	/* buffer locked by someone else */
155 	kstat_named_t	bio_bufdup;	/* duplicate buffer found for block */
156 };
157 
158 /*
159  * These flags are kept in b_flags.
160  * The first group is part of the DDI
161  */
162 #define	B_BUSY		0x0001	/* not on av_forw/back list */
163 #define	B_DONE		0x0002	/* transaction finished */
164 #define	B_ERROR		0x0004	/* transaction aborted */
165 #define	B_PAGEIO	0x0010	/* do I/O to pages on bp->p_pages */
166 #define	B_PHYS		0x0020	/* Physical IO potentially using UNIBUS map */
167 #define	B_READ		0x0040	/* read when I/O occurs */
168 #define	B_WRITE		0x0100	/* non-read pseudo-flag */
169 
170 /* Not part of the DDI */
171 #define	B_WANTED	0x0080		/* issue wakeup when BUSY goes off */
172 #define	B_AGE		0x000200	/* delayed write for correct aging */
173 #define	B_ASYNC		0x000400	/* don't wait for I/O completion */
174 #define	B_DELWRI	0x000800	/* delayed write-wait til buf needed */
175 #define	B_STALE		0x001000	/* on av_* list; invalid contents */
176 #define	B_DONTNEED	0x002000	/* after write, need not be cached */
177 #define	B_REMAPPED	0x004000	/* buffer is kernel addressable */
178 #define	B_FREE		0x008000	/* free page when done */
179 #define	B_INVAL		0x010000	/* destroy page when done */
180 #define	B_FORCE		0x020000	/* semi-permanent removal from cache */
181 #define	B_NOCACHE	0x080000 	/* don't cache block when released */
182 #define	B_TRUNC		0x100000	/* truncate page without I/O */
183 #define	B_SHADOW	0x200000	/* is b_shadow field valid? */
184 #define	B_RETRYWRI	0x400000	/* retry write til works or bfinval */
185 #define	B_FAILFAST	0x1000000	/* Fail promptly if device goes away */
186 #define	B_STARTED	0x2000000	/* io:::start probe called for buf */
187 #define	B_ABRWRITE	0x4000000	/* Application based recovery active */
188 #define	B_PAGE_NOWAIT	0x8000000	/* Skip the page if it is locked */
189 
190 /*
191  * There is some confusion over the meaning of B_FREE and B_INVAL and what
192  * the use of one over the other implies.
193  *
194  * In both cases, when we are done with the page (buffer) we want to free
195  * up the page.  In the case of B_FREE, the page will go to the cachelist.
196  * In the case of B_INVAL, the page will be destroyed (hashed out of it's
197  * vnode) and placed on the freelist.  Beyond this, there is no difference
198  * between the sole use of these two flags.  In both cases, IO will be done
199  * if the page is not yet committed to storage.
200  *
201  * In order to discard pages without writing them back, (B_INVAL | B_TRUNC)
202  * should be used.
203  *
204  * Use (B_INVAL | B_FORCE) to force the page to be destroyed even if we
205  * could not successfuly write out the page.
206  */
207 
208 /*
209  * Insq/Remq for the buffer hash lists.
210  */
211 #define	bremhash(bp) { \
212 	ASSERT((bp)->b_forw != NULL); \
213 	ASSERT((bp)->b_back != NULL); \
214 	(bp)->b_back->b_forw = (bp)->b_forw; \
215 	(bp)->b_forw->b_back = (bp)->b_back; \
216 	(bp)->b_forw = (bp)->b_back = NULL; \
217 }
218 #define	binshash(bp, dp) { \
219 	ASSERT((bp)->b_forw == NULL); \
220 	ASSERT((bp)->b_back == NULL); \
221 	ASSERT((dp)->b_forw != NULL); \
222 	ASSERT((dp)->b_back != NULL); \
223 	(bp)->b_forw = (dp)->b_forw; \
224 	(bp)->b_back = (dp); \
225 	(dp)->b_forw->b_back = (bp); \
226 	(dp)->b_forw = (bp); \
227 }
228 
229 
230 /*
231  * The hash structure maintains two lists:
232  *
233  * 	1) The hash list of buffers (b_forw & b_back)
234  *	2) The LRU free list of buffers on this hash bucket (av_forw & av_back)
235  *
236  * The dwbuf structure keeps a list of delayed write buffers per hash bucket
237  * hence there are exactly the same number of dwbuf structures as there are
238  * the hash buckets (hbuf structures) in the system.
239  *
240  * The number of buffers on the freelist may not be equal to the number of
241  * buffers on the hash list. That is because when buffers are busy they are
242  * taken off the freelist but not off the hash list. "b_length" field keeps
243  * track of the number of free buffers (including delayed writes ones) on
244  * the hash bucket. The "b_lock" mutex protects the free list as well as
245  * the hash list. It also protects the counter "b_length".
246  *
247  * Enties b_forw, b_back, av_forw & av_back must be at the same offset
248  * as the ones in buf structure.
249  */
250 struct	hbuf {
251 	int	b_flags;
252 
253 	struct	buf	*b_forw;	/* hash list forw pointer */
254 	struct	buf	*b_back;	/* hash list back pointer */
255 
256 	struct	buf	*av_forw;	/* free list forw pointer */
257 	struct	buf	*av_back;	/* free list back pointer */
258 
259 	int		b_length;	/* # of entries on free list */
260 	kmutex_t	b_lock;		/* lock to protect this structure */
261 };
262 
263 
264 /*
265  * The delayed list pointer entries should match with the buf strcuture.
266  */
267 struct	dwbuf {
268 	int	b_flags;		/* not used */
269 
270 	struct	buf	*b_forw;	/* not used */
271 	struct	buf	*b_back;	/* not used */
272 
273 	struct	buf	*av_forw;	/* delayed write forw pointer */
274 	struct	buf	*av_back;	/* delayed write back pointer */
275 };
276 
277 
278 /*
279  * Unlink a buffer from the available (free or delayed write) list and mark
280  * it busy (internal interface).
281  */
282 #define	notavail(bp) \
283 {\
284 	ASSERT(SEMA_HELD(&bp->b_sem)); \
285 	ASSERT((bp)->av_forw != NULL); \
286 	ASSERT((bp)->av_back != NULL); \
287 	ASSERT((bp)->av_forw != (bp)); \
288 	ASSERT((bp)->av_back != (bp)); \
289 	(bp)->av_back->av_forw = (bp)->av_forw; \
290 	(bp)->av_forw->av_back = (bp)->av_back; \
291 	(bp)->b_flags |= B_BUSY; \
292 	(bp)->av_forw = (bp)->av_back = NULL; \
293 }
294 
295 #if defined(_KERNEL)
296 /*
297  * Macros to avoid the extra function call needed for binary compat.
298  *
299  * B_RETRYWRI is not included in clear_flags for BWRITE(), BWRITE2(),
300  * or brwrite() so that the retry operation is persistent until the
301  * write either succeeds or the buffer is bfinval()'d.
302  *
303  */
304 #define	BREAD(dev, blkno, bsize) \
305 	bread_common(/* ufsvfsp */ NULL, dev, blkno, bsize)
306 
307 #define	BWRITE(bp) \
308 	bwrite_common(/* ufsvfsp */ NULL, bp, /* force_wait */ 0, \
309 		/* do_relse */ 1, \
310 		/* clear_flags */ (B_READ | B_DONE | B_ERROR | B_DELWRI))
311 
312 #define	BWRITE2(bp) \
313 	bwrite_common(/* ufsvfsp */ NULL, bp, /* force_wait */ 1, \
314 		/* do_relse */ 0, \
315 		/* clear_flags */ (B_READ | B_DONE | B_ERROR | B_DELWRI))
316 
317 #define	GETBLK(dev, blkno, bsize) \
318 	getblk_common(/* ufsvfsp */ NULL, dev, blkno, bsize, /* errflg */ 0)
319 
320 
321 /*
322  * Macros for new retry write interfaces.
323  */
324 
325 /*
326  * Same as bdwrite() except write failures are retried.
327  */
328 #define	bdrwrite(bp) { \
329 	(bp)->b_flags |= B_RETRYWRI; \
330 	bdwrite((bp)); \
331 }
332 
333 /*
334  * Same as bwrite() except write failures are retried.
335  */
336 #define	brwrite(bp) { \
337 	(bp)->b_flags |= B_RETRYWRI; \
338 	bwrite_common((bp), /* force_wait */ 0, /* do_relse */ 1, \
339 		/* clear_flags */ (B_READ | B_DONE | B_ERROR | B_DELWRI)); \
340 }
341 
342 extern struct hbuf	*hbuf;		/* Hash table */
343 extern struct dwbuf	*dwbuf;		/* delayed write hash table */
344 extern struct buf	*buf;		/* The buffer pool itself */
345 extern struct buf	bfreelist;	/* head of available list */
346 
347 extern void (*bio_lufs_strategy)(void *, buf_t *);	/* UFS Logging */
348 extern void (*bio_snapshot_strategy)(void *, buf_t *);	/* UFS snapshots */
349 
350 int	bcheck(dev_t, struct buf *);
351 int	iowait(struct buf *);
352 int	hash2ints(int x, int y);
353 int	bio_busy(int);
354 int	biowait(struct buf *);
355 int	biomodified(struct buf *);
356 int	geterror(struct buf *);
357 void	minphys(struct buf *);
358 /*
359  * ufsvfsp is declared as a void * to avoid having everyone that uses
360  * this header file include sys/fs/ufs_inode.h.
361  */
362 void	bwrite_common(void *ufsvfsp, struct buf *, int force_wait,
363 	int do_relse, int clear_flags);
364 void	bwrite(struct buf *);
365 void	bwrite2(struct buf *);
366 void	bdwrite(struct buf *);
367 void	bawrite(struct buf *);
368 void	brelse(struct buf *);
369 void	iodone(struct buf *);
370 void	clrbuf(struct buf *);
371 void	bflush(dev_t);
372 void	blkflush(dev_t, daddr_t);
373 void	binval(dev_t);
374 int	bfinval(dev_t, int);
375 void	binit(void);
376 void	biodone(struct buf *);
377 void	bioinit(struct buf *);
378 void	biofini(struct buf *);
379 void	bp_mapin(struct buf *);
380 void	*bp_mapin_common(struct buf *, int);
381 void	bp_mapout(struct buf *);
382 int	bp_copyin(struct buf *, void *, offset_t, size_t);
383 int	bp_copyout(void *, struct buf *, offset_t, size_t);
384 void	bp_init(size_t, uint_t);
385 int	bp_color(struct buf *);
386 void	pageio_done(struct buf *);
387 struct buf *bread(dev_t, daddr_t, long);
388 struct buf *bread_common(void *, dev_t, daddr_t, long);
389 struct buf *breada(dev_t, daddr_t, daddr_t, long);
390 struct buf *getblk(dev_t, daddr_t, long);
391 struct buf *getblk_common(void *, dev_t, daddr_t, long, int);
392 struct buf *ngeteblk(long);
393 struct buf *geteblk(void);
394 struct buf *pageio_setup(struct page *, size_t, struct vnode *, int);
395 void bioerror(struct buf *bp, int error);
396 void bioreset(struct buf *bp);
397 struct buf *bioclone(struct buf *, off_t, size_t, dev_t, daddr_t,
398 	int (*)(struct buf *), struct buf *, int);
399 size_t	biosize(void);
400 #endif	/* defined(_KERNEL) */
401 
402 #ifdef	__cplusplus
403 }
404 #endif
405 
406 #endif	/* _SYS_BUF_H */
407