2 * Copyright (c) 1982, 1986, 1989, 1993
3 * The Regents of the University of California. All rights reserved.
4 * (c) UNIX System Laboratories, Inc.
5 * All or some portions of this file are derived from material licensed
6 * to the University of California by American Telephone and Telegraph
7 * Co. or Unix System Laboratories, Inc. and are reproduced herein with
8 * the permission of UNIX System Laboratories, Inc.
10 * Redistribution and use in source and binary forms, with or without
11 * modification, are permitted provided that the following conditions
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 * 3. All advertising materials mentioning features or use of this software
19 * must display the following acknowledgement:
20 * This product includes software developed by the University of
21 * California, Berkeley and its contributors.
22 * 4. Neither the name of the University nor the names of its contributors
23 * may be used to endorse or promote products derived from this software
24 * without specific prior written permission.
26 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
27 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
28 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
29 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
30 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
31 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
32 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
33 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
34 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
35 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
38 * @(#)buf.h 8.9 (Berkeley) 3/30/95
39 * $FreeBSD: src/sys/sys/buf.h,v 1.88.2.10 2003/01/25 19:02:23 dillon Exp $
40 * $DragonFly: src/sys/sys/buf.h,v 1.34 2006/05/03 20:44:49 dillon Exp $
47 #include <sys/queue.h>
52 #ifndef _SYS_DEVICE_H_
53 #include <sys/device.h>
65 #ifndef _SYS_SPINLOCK_H_
66 #include <sys/spinlock.h>
79 RB_PROTOTYPE2(buf_rb_tree, buf, b_rbnode, rb_buf_compare, off_t);
80 RB_PROTOTYPE2(buf_rb_hash, buf, b_rbhash, rb_buf_compare, off_t);
83 * To avoid including <ufs/ffs/softdep.h>
85 LIST_HEAD(workhead, worklist);
87 * These are currently used only by the soft dependency code, hence
88 * are stored once in a global variable. If other subsystems wanted
89 * to use these hooks, a pointer to a set of bio_ops could be added
92 extern struct bio_ops {
93 void (*io_start) (struct buf *);
94 void (*io_complete) (struct buf *);
95 void (*io_deallocate) (struct buf *);
96 int (*io_fsync) (struct vnode *);
97 int (*io_sync) (struct mount *);
98 void (*io_movedeps) (struct buf *, struct buf *);
99 int (*io_countdeps) (struct buf *, int);
102 typedef enum buf_cmd {
111 * The buffer header describes an I/O operation in the kernel.
114 * b_bufsize represents the filesystem block size (for this particular
115 * block) and/or the allocation size or original request size. This
116 * field is NOT USED by lower device layers. VNode and device
117 * strategy routines WILL NEVER ACCESS THIS FIELD.
119 * b_bcount represents the EOF-clipped request size. It is typically
120 * set to b_bufsize prior to I/O initiation and may be modified by
121 * the driver chain (for example, to clip upon encountering the end
122 * of the block device). b_bcount may only be clipped to represent
123 * EOF - for example, it would be clipped to the symlink length when
124 * reading a symlink, or to the file EOF. It is never clipped due to
125 * an error, nor is it clipped on a zero-fill short read. For byte
126 * oriented files b_bcount is typically set to b_bufsize to initiate
127 * the read or write to the underlying block device, then clipped to
128 * the file EOF upon completion of the read or write.
130 * b_resid. Number of bytes remaining in I/O. After an I/O operation
131 * completes, b_resid is usually 0 indicating 100% success. Note however
132 * that if the device chain encounters an EOF, both b_resid and b_bcount
133 * will be truncated. So b_resid will also be 0 if a short-read (EOF)
134 * occurs and the caller must check for the EOF condition by comparing
135 * b_bcount against (typically) b_bufsize.
137 * b_dirtyoff, b_dirtyend. Buffers support piecemeal, unaligned
138 * ranges of dirty data that need to be written to backing store.
139 * The range is typically clipped at b_bcount (not b_bufsize).
141 * b_bio1 and b_bio2 represent the two primary I/O layers. Additional
142 * I/O layers are allocated out of the object cache and may also exist.
144 * b_bio1 is the logical layer and contains offset or block number
145 * data for the primary vnode, b_vp. I/O operations are almost
146 * universally initiated from the logical layer, so you will often
147 * see things like: vn_strategy(bp->b_vp, &bp->b_bio1).
149 * b_bio2 is the first physical layer (typically the slice-relative
150 * layer) and contains the translated offset or block number for
151 * the block device underlying a filesystem. Filesystems such as UFS
152 * will maintain cached translations and you may see them initiate
153 * a 'physical' I/O using vn_strategy(devvp, &bp->b_bio2). BUT,
154 * remember that the layering is relative to bp->b_vp, so the
155 * device-relative block numbers for buffer cache operations that occur
156 * directly on a block device will be in the first BIO layer.
158 * NOTE!!! Only the BIO subsystem accesses b_bio1 and b_bio2 directly.
159 * ALL STRATEGY LAYERS FOR BOTH VNODES AND DEVICES ONLY ACCESS THE BIO
160 * PASSED TO THEM, AND WILL PUSH ANOTHER BIO LAYER IF FORWARDING THE
161 * I/O DEEPER. In particular, a vn_strategy() or dev_dstrategy()
162 * call should not ever access buf->b_vp as this vnode may be totally
163 * unrelated to the vnode/device whos strategy routine was called.
166 RB_ENTRY(buf) b_rbnode; /* RB node in vnode clean/dirty tree */
167 RB_ENTRY(buf) b_rbhash; /* RB node in vnode hash tree */
168 TAILQ_ENTRY(buf) b_freelist; /* Free list position if not active. */
169 struct buf *b_cluster_next; /* Next buffer (cluster code) */
170 struct vnode *b_vp; /* (vp, loffset) index */
171 struct bio b_bio_array[NBUF_BIO]; /* BIO translation layers */
172 u_int32_t b_flags; /* B_* flags. */
173 unsigned short b_qindex; /* buffer queue index */
174 unsigned short b_unused01;
175 struct lock b_lock; /* Buffer lock */
176 buf_cmd_t b_cmd; /* I/O command */
177 int b_bufsize; /* Allocated buffer size. */
178 int b_runningbufspace; /* when I/O is running, pipelining */
179 int b_bcount; /* Valid bytes in buffer. */
180 int b_resid; /* Remaining I/O */
181 int b_error; /* Error return */
182 caddr_t b_data; /* Memory, superblocks, indirect etc. */
183 caddr_t b_kvabase; /* base kva for buffer */
184 int b_kvasize; /* size of kva for buffer */
185 int b_dirtyoff; /* Offset in buffer of dirty region. */
186 int b_dirtyend; /* Offset of end of dirty region. */
187 struct xio b_xio; /* data buffer page list management */
188 struct workhead b_dep; /* List of filesystem dependencies. */
194 #define b_bio1 b_bio_array[0] /* logical layer */
195 #define b_bio2 b_bio_array[1] /* (typically) the disk layer */
196 #define b_loffset b_bio1.bio_offset
199 * These flags are kept in b_flags.
203 * B_ASYNC VOP calls on bp's are usually async whether or not
204 * B_ASYNC is set, but some subsystems, such as NFS, like
205 * to know what is best for the caller so they can
208 * B_PAGING Indicates that bp is being used by the paging system or
209 * some paging system and that the bp is not linked into
210 * the b_vp's clean/dirty linked lists or ref counts.
211 * Buffer vp reassignments are illegal in this case.
213 * B_CACHE This may only be set if the buffer is entirely valid.
214 * The situation where B_DELWRI is set and B_CACHE is
215 * clear MUST be committed to disk by getblk() so
216 * B_DELWRI can also be cleared. See the comments for
217 * getblk() in kern/vfs_bio.c. If B_CACHE is clear,
218 * the caller is expected to clear B_ERROR|B_INVAL,
219 * set BUF_CMD_READ, and initiate an I/O.
221 * The 'entire buffer' is defined to be the range from
222 * 0 through b_bcount.
224 * B_MALLOC Request that the buffer be allocated from the malloc
225 * pool, DEV_BSIZE aligned instead of PAGE_SIZE aligned.
227 * B_CLUSTEROK This flag is typically set for B_DELWRI buffers
228 * by filesystems that allow clustering when the buffer
229 * is fully dirty and indicates that it may be clustered
230 * with other adjacent dirty buffers. Note the clustering
231 * may not be used with the stage 1 data write under NFS
232 * but may be used for the commit rpc portion.
234 * B_VMIO Indicates that the buffer is tied into an VM object.
235 * The buffer's data is always PAGE_SIZE aligned even
236 * if b_bufsize and b_bcount are not. ( b_bufsize is
237 * always at least DEV_BSIZE aligned, though ).
239 * B_DIRECT Hint that we should attempt to completely free
240 * the pages underlying the buffer. B_DIRECT is
241 * sticky until the buffer is released and typically
242 * only has an effect when B_RELBUF is also set.
244 * B_NOWDRAIN This flag should be set when a device (like VN)
245 * does a turn-around VOP_WRITE from its strategy
246 * routine. This flag prevents bwrite() from blocking
247 * in wdrain, avoiding a deadlock situation.
250 #define B_AGE 0x00000001 /* Move to age queue when I/O done. */
251 #define B_NEEDCOMMIT 0x00000002 /* Append-write in progress. */
252 #define B_ASYNC 0x00000004 /* Start I/O, do not wait. */
253 #define B_DIRECT 0x00000008 /* direct I/O flag (pls free vmio) */
254 #define B_DEFERRED 0x00000010 /* Skipped over for cleaning */
255 #define B_CACHE 0x00000020 /* Bread found us in the cache. */
256 #define B_HASHED 0x00000040 /* Indexed via v_rbhash_tree */
257 #define B_DELWRI 0x00000080 /* Delay I/O until buffer reused. */
258 #define B_UNUSED0100 0x00000100
259 #define B_UNUSED0200 0x00000200
260 #define B_EINTR 0x00000400 /* I/O was interrupted */
261 #define B_ERROR 0x00000800 /* I/O error occurred. */
262 #define B_UNUSED1000 0x00001000 /* Unused */
263 #define B_INVAL 0x00002000 /* Does not contain valid info. */
264 #define B_LOCKED 0x00004000 /* Locked in core (not reusable). */
265 #define B_NOCACHE 0x00008000 /* Do not cache block after use. */
266 #define B_MALLOC 0x00010000 /* malloced b_data */
267 #define B_CLUSTEROK 0x00020000 /* Pagein op, so swap() can count it. */
268 #define B_UNUSED40000 0x00040000
269 #define B_RAW 0x00080000 /* Set by physio for raw transfers. */
270 #define B_UNUSED100000 0x00100000
271 #define B_DIRTY 0x00200000 /* Needs writing later. */
272 #define B_RELBUF 0x00400000 /* Release VMIO buffer. */
273 #define B_WANT 0x00800000 /* Used by vm_pager.c */
274 #define B_VNCLEAN 0x01000000 /* On vnode clean list */
275 #define B_VNDIRTY 0x02000000 /* On vnode dirty list */
276 #define B_PAGING 0x04000000 /* volatile paging I/O -- bypass VMIO */
277 #define B_ORDERED 0x08000000 /* Must guarantee I/O ordering */
278 #define B_RAM 0x10000000 /* Read ahead mark (flag) */
279 #define B_VMIO 0x20000000 /* VMIO flag */
280 #define B_CLUSTER 0x40000000 /* pagein op, so swap() can count it */
281 #define B_NOWDRAIN 0x80000000 /* Avoid wdrain deadlock */
283 #define PRINT_BUF_FLAGS "\20" \
284 "\40nowdrain\37cluster\36vmio\35ram\34ordered" \
285 "\33paging\32vndirty\31vnclean\30want\27relbuf\26dirty" \
286 "\25unused20\24raw\23unused18\22clusterok\21malloc\20nocache" \
287 "\17locked\16inval\15unused12\14error\13eintr\12unused9\11unused8" \
288 "\10delwri\7hashed\6cache\5deferred\4direct\3async\2needcommit\1age"
290 #define NOOFFSET (-1LL) /* No buffer offset calculated yet */
294 * Buffer locking. See sys/buf2.h for inline functions.
296 extern char *buf_wmesg; /* Default buffer lock message */
297 #define BUF_WMESG "bufwait"
301 struct bio_queue_head {
302 TAILQ_HEAD(bio_queue, bio) queue;
304 struct bio *insert_point;
305 struct bio *switch_point;
309 * This structure describes a clustered I/O.
311 struct cluster_save {
312 int bs_nchildren; /* Number of associated buffers. */
313 struct buf **bs_children; /* List of associated buffers. */
317 * Zero out the buffer's data area.
319 #define clrbuf(bp) { \
320 bzero((bp)->b_data, (u_int)(bp)->b_bcount); \
325 * Flags to low-level bitmap allocation routines (balloc).
327 * Note: sequential_heuristic() in kern/vfs_vnops.c limits the count
330 #define B_SEQMASK 0x7F000000 /* Sequential heuristic mask. */
331 #define B_SEQSHIFT 24 /* Sequential heuristic shift. */
332 #define B_SEQMAX 0x7F
333 #define B_CLRBUF 0x01 /* Cleared invalid areas of buffer. */
334 #define B_SYNC 0x02 /* Do all allocations synchronously. */
337 extern int nbuf; /* The number of buffer headers */
338 extern int maxswzone; /* Max KVA for swap structures */
339 extern int maxbcache; /* Max KVA for buffer cache */
340 extern int runningbufspace;
341 extern int buf_maxio; /* nominal maximum I/O for buffer */
342 extern struct buf *buf; /* The buffer headers. */
343 extern char *buffers; /* The buffer contents. */
344 extern int bufpages; /* Number of memory pages in the buffer pool. */
345 extern struct buf *swbuf; /* Swap I/O buffer headers. */
346 extern int nswbuf; /* Number of swap I/O buffer headers. */
351 void bwillwrite (void);
352 int buf_dirty_count_severe (void);
353 void initbufbio(struct buf *);
354 void reinitbufbio(struct buf *);
355 void clearbiocache(struct bio *);
356 void bremfree (struct buf *);
357 int bread (struct vnode *, off_t, int, struct buf **);
358 int breadn (struct vnode *, off_t, int, off_t *, int *, int,
360 int bwrite (struct buf *);
361 void bdwrite (struct buf *);
362 void bawrite (struct buf *);
363 void bdirty (struct buf *);
364 void bundirty (struct buf *);
365 int bowrite (struct buf *);
366 void brelse (struct buf *);
367 void bqrelse (struct buf *);
368 int vfs_bio_awrite (struct buf *);
369 struct buf *getpbuf (int *);
370 int inmem (struct vnode *, off_t);
371 struct buf *findblk (struct vnode *, off_t);
372 struct buf *getblk (struct vnode *, off_t, int, int, int);
373 struct buf *geteblk (int);
374 struct bio *push_bio(struct bio *);
375 void pop_bio(struct bio *);
376 int biowait (struct buf *);
377 void biodone (struct bio *);
379 void cluster_append(struct bio *, struct buf *);
380 int cluster_read (struct vnode *, off_t, off_t, int,
381 int, int, struct buf **);
382 int cluster_wbuild (struct vnode *, int, off_t, int);
383 void cluster_write (struct buf *, off_t, int);
384 int physio (dev_t dev, struct uio *uio, int ioflag);
385 #define physread physio
386 #define physwrite physio
387 void vfs_bio_set_validclean (struct buf *, int base, int size);
388 void vfs_bio_clrbuf (struct buf *);
389 void vfs_busy_pages (struct vnode *, struct buf *);
390 void vfs_unbusy_pages (struct buf *);
391 int vmapbuf (struct buf *, caddr_t, int);
392 void vunmapbuf (struct buf *);
393 void relpbuf (struct buf *, int *);
394 void brelvp (struct buf *);
395 void bgetvp (struct vnode *, struct buf *);
396 int allocbuf (struct buf *bp, int size);
397 int scan_all_buffers (int (*)(struct buf *, void *), void *);
398 void reassignbuf (struct buf *);
399 struct buf *trypbuf (int *);
403 #endif /* !_SYS_BUF_H_ */