Replace the the buffer cache's B_READ, B_WRITE, B_FORMAT, and B_FREEBUF
[dragonfly.git] / sys / sys / buf.h
CommitLineData
984263bc
MD
1/*
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.
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 * 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.
25 *
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
36 * SUCH DAMAGE.
37 *
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 $
10f3fee5 40 * $DragonFly: src/sys/sys/buf.h,v 1.31 2006/04/30 17:22:17 dillon Exp $
984263bc
MD
41 */
42
43#ifndef _SYS_BUF_H_
44#define _SYS_BUF_H_
45
335dda38 46#ifndef _SYS_QUEUE_H_
984263bc 47#include <sys/queue.h>
335dda38
MD
48#endif
49#ifndef _SYS_LOCK_H_
984263bc 50#include <sys/lock.h>
335dda38
MD
51#endif
52#ifndef _SYS_DEVICE_H_
53#include <sys/device.h>
54#endif
984263bc 55
54f51aeb
HP
56#ifndef _SYS_XIO_H_
57#include <sys/xio.h>
58#endif
6bae6177
MD
59#ifndef _SYS_TREE_H_
60#include <sys/tree.h>
61#endif
1eabbf32
HP
62#ifndef _SYS_BIO_H_
63#include <sys/bio.h>
64#endif
16523a43
MD
65#ifndef _SYS_SPINLOCK_H_
66#include <sys/spinlock.h>
67#endif
1eabbf32 68
984263bc 69struct buf;
81b5c339 70struct bio;
984263bc
MD
71struct mount;
72struct vnode;
54f51aeb 73struct xio;
984263bc 74
81b5c339
MD
75#define NBUF_BIO 4
76
6bae6177 77struct buf_rb_tree;
1f1ea522 78struct buf_rb_hash;
e656fe10
MD
79RB_PROTOTYPE2(buf_rb_tree, buf, b_rbnode, rb_buf_compare, off_t);
80RB_PROTOTYPE2(buf_rb_hash, buf, b_rbhash, rb_buf_compare, off_t);
6bae6177 81
984263bc
MD
82/*
83 * To avoid including <ufs/ffs/softdep.h>
84 */
85LIST_HEAD(workhead, worklist);
86/*
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
90 * to each buffer.
91 */
92extern struct bio_ops {
b153f746
RG
93 void (*io_start) (struct buf *);
94 void (*io_complete) (struct buf *);
95 void (*io_deallocate) (struct buf *);
1e5a0c40
HP
96 int (*io_fsync) (struct vnode *);
97 int (*io_sync) (struct mount *);
b153f746 98 void (*io_movedeps) (struct buf *, struct buf *);
1e5a0c40 99 int (*io_countdeps) (struct buf *, int);
984263bc
MD
100} bioops;
101
10f3fee5
MD
102typedef enum buf_cmd {
103 BUF_CMD_DONE = 0,
104 BUF_CMD_READ,
105 BUF_CMD_WRITE,
106 BUF_CMD_FREEBLKS,
107 BUF_CMD_FORMAT
108} buf_cmd_t;
109
984263bc
MD
110/*
111 * The buffer header describes an I/O operation in the kernel.
112 *
113 * NOTES:
114 * b_bufsize, b_bcount. b_bufsize is the allocation size of the
115 * buffer, either DEV_BSIZE or PAGE_SIZE aligned. b_bcount is the
116 * originally requested buffer size and can serve as a bounds check
117 * against EOF. For most, but not all uses, b_bcount == b_bufsize.
118 *
119 * b_dirtyoff, b_dirtyend. Buffers support piecemeal, unaligned
120 * ranges of dirty data that need to be written to backing store.
121 * The range is typically clipped at b_bcount ( not b_bufsize ).
122 *
123 * b_resid. Number of bytes remaining in I/O. After an I/O operation
124 * completes, b_resid is usually 0 indicating 100% success.
81b5c339
MD
125 *
126 * b_bio1 and b_bio2 represent the two primary I/O layers. Additional
127 * I/O layers are allocated out of the object cache and may also exist.
128 *
129 * b_bio1 is the logical layer and contains offset or block number
130 * data for the primary vnode, b_vp. I/O operations are almost
131 * universally initiated from the logical layer, so you will often
132 * see things like: vn_strategy(bp->b_vp, &bp->b_bio1).
133 *
134 * b_bio2 is the first physical layer (typically the slice-relative
135 * layer) and contains the translated offset or block number for
136 * the block device underlying a filesystem. Filesystems such as UFS
137 * will maintain cached translations and you may see them initiate
138 * a 'physical' I/O using vn_strategy(devvp, &bp->b_bio2). BUT,
139 * remember that the layering is relative to bp->b_vp, so the
140 * device-relative block numbers for buffer cache operations that occur
141 * directly on a block device will be in the first BIO layer.
142 *
143 * NOTE!!! Only the BIO subsystem accesses b_bio1 and b_bio2 directly.
144 * ALL STRATEGY LAYERS FOR BOTH VNODES AND DEVICES ONLY ACCESS THE BIO
145 * PASSED TO THEM, AND WILL PUSH ANOTHER BIO LAYER IF FORWARDING THE
146 * I/O DEEPER. In particular, a vn_strategy() or dev_dstrategy()
147 * call should not ever access buf->b_vp as this vnode may be totally
148 * unrelated to the vnode/device whos strategy routine was called.
984263bc
MD
149 */
150struct buf {
1f1ea522
MD
151 RB_ENTRY(buf) b_rbnode; /* RB node in vnode clean/dirty tree */
152 RB_ENTRY(buf) b_rbhash; /* RB node in vnode hash tree */
984263bc 153 TAILQ_ENTRY(buf) b_freelist; /* Free list position if not active. */
81b5c339 154 struct buf *b_cluster_next; /* Next buffer (cluster code) */
54078292 155 struct vnode *b_vp; /* (vp, loffset) index */
81b5c339 156 struct bio b_bio_array[NBUF_BIO]; /* BIO translation layers */
54078292 157 u_int32_t b_flags; /* B_* flags. */
984263bc
MD
158 unsigned short b_qindex; /* buffer queue index */
159 unsigned char b_xflags; /* extra flags */
54078292 160 unsigned char b_unused01;
984263bc 161 struct lock b_lock; /* Buffer lock */
10f3fee5 162 buf_cmd_t b_cmd; /* I/O command */
54078292
MD
163 int b_bufsize; /* Allocated buffer size. */
164 int b_runningbufspace; /* when I/O is running, pipelining */
165 int b_bcount; /* Valid bytes in buffer. */
166 int b_resid; /* Remaining I/O */
81b5c339 167 int b_error; /* Error return */
984263bc
MD
168 caddr_t b_data; /* Memory, superblocks, indirect etc. */
169 caddr_t b_kvabase; /* base kva for buffer */
170 int b_kvasize; /* size of kva for buffer */
984263bc
MD
171 int b_dirtyoff; /* Offset in buffer of dirty region. */
172 int b_dirtyend; /* Offset of end of dirty region. */
984263bc 173 void *b_saveaddr; /* Original b_addr for physio. */
81b5c339 174 struct xio b_xio; /* data buffer page list management */
984263bc 175 struct workhead b_dep; /* List of filesystem dependencies. */
984263bc
MD
176};
177
81b5c339
MD
178/*
179 * XXX temporary
180 */
181#define b_bio1 b_bio_array[0] /* logical layer */
182#define b_bio2 b_bio_array[1] /* (typically) the disk layer */
183#define b_loffset b_bio1.bio_offset
984263bc
MD
184
185/*
186 * These flags are kept in b_flags.
187 *
188 * Notes:
189 *
190 * B_ASYNC VOP calls on bp's are usually async whether or not
191 * B_ASYNC is set, but some subsystems, such as NFS, like
192 * to know what is best for the caller so they can
193 * optimize the I/O.
194 *
195 * B_PAGING Indicates that bp is being used by the paging system or
196 * some paging system and that the bp is not linked into
197 * the b_vp's clean/dirty linked lists or ref counts.
198 * Buffer vp reassignments are illegal in this case.
199 *
200 * B_CACHE This may only be set if the buffer is entirely valid.
201 * The situation where B_DELWRI is set and B_CACHE is
202 * clear MUST be committed to disk by getblk() so
203 * B_DELWRI can also be cleared. See the comments for
204 * getblk() in kern/vfs_bio.c. If B_CACHE is clear,
205 * the caller is expected to clear B_ERROR|B_INVAL,
10f3fee5 206 * set BUF_CMD_READ, and initiate an I/O.
984263bc
MD
207 *
208 * The 'entire buffer' is defined to be the range from
209 * 0 through b_bcount.
210 *
211 * B_MALLOC Request that the buffer be allocated from the malloc
212 * pool, DEV_BSIZE aligned instead of PAGE_SIZE aligned.
213 *
214 * B_CLUSTEROK This flag is typically set for B_DELWRI buffers
215 * by filesystems that allow clustering when the buffer
216 * is fully dirty and indicates that it may be clustered
217 * with other adjacent dirty buffers. Note the clustering
218 * may not be used with the stage 1 data write under NFS
219 * but may be used for the commit rpc portion.
220 *
221 * B_VMIO Indicates that the buffer is tied into an VM object.
222 * The buffer's data is always PAGE_SIZE aligned even
223 * if b_bufsize and b_bcount are not. ( b_bufsize is
224 * always at least DEV_BSIZE aligned, though ).
225 *
226 * B_DIRECT Hint that we should attempt to completely free
227 * the pages underlying the buffer. B_DIRECT is
228 * sticky until the buffer is released and typically
229 * only has an effect when B_RELBUF is also set.
230 *
231 * B_NOWDRAIN This flag should be set when a device (like VN)
232 * does a turn-around VOP_WRITE from its strategy
233 * routine. This flag prevents bwrite() from blocking
234 * in wdrain, avoiding a deadlock situation.
235 */
236
237#define B_AGE 0x00000001 /* Move to age queue when I/O done. */
238#define B_NEEDCOMMIT 0x00000002 /* Append-write in progress. */
239#define B_ASYNC 0x00000004 /* Start I/O, do not wait. */
240#define B_DIRECT 0x00000008 /* direct I/O flag (pls free vmio) */
241#define B_DEFERRED 0x00000010 /* Skipped over for cleaning */
242#define B_CACHE 0x00000020 /* Bread found us in the cache. */
1f1ea522 243#define B_HASHED 0x00000040 /* Indexed via v_rbhash_tree */
984263bc 244#define B_DELWRI 0x00000080 /* Delay I/O until buffer reused. */
10f3fee5
MD
245#define B_UNUSED0100 0x00000100
246#define B_UNUSED0200 0x00000200
984263bc
MD
247#define B_EINTR 0x00000400 /* I/O was interrupted */
248#define B_ERROR 0x00000800 /* I/O error occurred. */
71bc1d95 249#define B_UNUSED1000 0x00001000 /* Unused */
984263bc
MD
250#define B_INVAL 0x00002000 /* Does not contain valid info. */
251#define B_LOCKED 0x00004000 /* Locked in core (not reusable). */
252#define B_NOCACHE 0x00008000 /* Do not cache block after use. */
253#define B_MALLOC 0x00010000 /* malloced b_data */
254#define B_CLUSTEROK 0x00020000 /* Pagein op, so swap() can count it. */
9fd739f4 255#define B_UNUSED40000 0x00040000
984263bc 256#define B_RAW 0x00080000 /* Set by physio for raw transfers. */
10f3fee5 257#define B_UNUSED100000 0x00100000
984263bc
MD
258#define B_DIRTY 0x00200000 /* Needs writing later. */
259#define B_RELBUF 0x00400000 /* Release VMIO buffer. */
260#define B_WANT 0x00800000 /* Used by vm_pager.c */
71bc1d95 261#define B_UNUSED1000000 0x01000000 /* Unused */
10f3fee5 262#define B_UNUSED2000000 0x02000000
984263bc
MD
263#define B_PAGING 0x04000000 /* volatile paging I/O -- bypass VMIO */
264#define B_ORDERED 0x08000000 /* Must guarantee I/O ordering */
265#define B_RAM 0x10000000 /* Read ahead mark (flag) */
266#define B_VMIO 0x20000000 /* VMIO flag */
267#define B_CLUSTER 0x40000000 /* pagein op, so swap() can count it */
268#define B_NOWDRAIN 0x80000000 /* Avoid wdrain deadlock */
269
270#define PRINT_BUF_FLAGS "\20\40nowdrain\37cluster\36vmio\35ram\34ordered" \
271 "\33paging\32xxx\31writeinprog\30want\27relbuf\26dirty" \
272 "\25read\24raw\23phys\22clusterok\21malloc\20nocache" \
273 "\17locked\16inval\15scanned\14error\13eintr\12done\11freebuf" \
274 "\10delwri\7call\6cache\4direct\3async\2needcommit\1age"
275
276/*
277 * These flags are kept in b_xflags.
278 */
279#define BX_VNDIRTY 0x00000001 /* On vnode dirty list */
280#define BX_VNCLEAN 0x00000002 /* On vnode clean list */
70371608
MD
281#define BX_UNUSED0004 0x00000004
282#define BX_UNUSED0008 0x00000008
283#define BX_UNUSED0010 0x00000010
984263bc
MD
284#define BX_AUTOCHAINDONE 0x00000020 /* pager I/O chain auto mode */
285
286#define NOOFFSET (-1LL) /* No buffer offset calculated yet */
287
288#ifdef _KERNEL
289/*
3020e3be 290 * Buffer locking. See sys/buf2.h for inline functions.
984263bc 291 */
984263bc
MD
292extern char *buf_wmesg; /* Default buffer lock message */
293#define BUF_WMESG "bufwait"
984263bc
MD
294
295#endif /* _KERNEL */
296
81b5c339
MD
297struct bio_queue_head {
298 TAILQ_HEAD(bio_queue, bio) queue;
54078292 299 off_t last_offset;
81b5c339
MD
300 struct bio *insert_point;
301 struct bio *switch_point;
984263bc
MD
302};
303
304/*
54078292 305 * This structure describes a clustered I/O.
984263bc
MD
306 */
307struct cluster_save {
984263bc
MD
308 int bs_nchildren; /* Number of associated buffers. */
309 struct buf **bs_children; /* List of associated buffers. */
310};
311
984263bc
MD
312/*
313 * Zero out the buffer's data area.
314 */
315#define clrbuf(bp) { \
316 bzero((bp)->b_data, (u_int)(bp)->b_bcount); \
317 (bp)->b_resid = 0; \
318}
319
320/*
321 * Flags to low-level bitmap allocation routines (balloc).
322 *
323 * Note: sequential_heuristic() in kern/vfs_vnops.c limits the count
324 * to 127.
325 */
326#define B_SEQMASK 0x7F000000 /* Sequential heuristic mask. */
327#define B_SEQSHIFT 24 /* Sequential heuristic shift. */
328#define B_SEQMAX 0x7F
329#define B_CLRBUF 0x01 /* Cleared invalid areas of buffer. */
330#define B_SYNC 0x02 /* Do all allocations synchronously. */
331
332#ifdef _KERNEL
333extern int nbuf; /* The number of buffer headers */
334extern int maxswzone; /* Max KVA for swap structures */
335extern int maxbcache; /* Max KVA for buffer cache */
336extern int runningbufspace;
337extern int buf_maxio; /* nominal maximum I/O for buffer */
41cbf46c 338extern struct buf *buf; /* The buffer headers. */
984263bc
MD
339extern char *buffers; /* The buffer contents. */
340extern int bufpages; /* Number of memory pages in the buffer pool. */
341extern struct buf *swbuf; /* Swap I/O buffer headers. */
342extern int nswbuf; /* Number of swap I/O buffer headers. */
984263bc
MD
343
344struct uio;
345
b153f746
RG
346void bufinit (void);
347void bwillwrite (void);
348int buf_dirty_count_severe (void);
81b5c339
MD
349void initbufbio(struct buf *);
350void reinitbufbio(struct buf *);
351void clearbiocache(struct bio *);
b153f746 352void bremfree (struct buf *);
54078292
MD
353int bread (struct vnode *, off_t, int, struct buf **);
354int breadn (struct vnode *, off_t, int, off_t *, int *, int,
b153f746
RG
355 struct buf **);
356int bwrite (struct buf *);
357void bdwrite (struct buf *);
358void bawrite (struct buf *);
359void bdirty (struct buf *);
360void bundirty (struct buf *);
361int bowrite (struct buf *);
362void brelse (struct buf *);
363void bqrelse (struct buf *);
364int vfs_bio_awrite (struct buf *);
81b5c339 365struct buf *getpbuf (int *);
54078292
MD
366int inmem (struct vnode *, off_t);
367struct buf *findblk (struct vnode *, off_t);
368struct buf *getblk (struct vnode *, off_t, int, int, int);
b153f746 369struct buf *geteblk (int);
81b5c339
MD
370struct bio *push_bio(struct bio *);
371void pop_bio(struct bio *);
b153f746 372int biowait (struct buf *);
81b5c339 373void biodone (struct bio *);
984263bc 374
81b5c339 375void cluster_append(struct bio *, struct buf *);
54078292
MD
376int cluster_read (struct vnode *, off_t, off_t, int,
377 int, int, struct buf **);
378int cluster_wbuild (struct vnode *, int, off_t, int);
379void cluster_write (struct buf *, off_t, int);
b153f746 380int physio (dev_t dev, struct uio *uio, int ioflag);
984263bc
MD
381#define physread physio
382#define physwrite physio
b153f746
RG
383void vfs_bio_set_validclean (struct buf *, int base, int size);
384void vfs_bio_clrbuf (struct buf *);
10f3fee5 385void vfs_busy_pages (struct vnode *, struct buf *);
b153f746 386void vfs_unbusy_pages (struct buf *);
b153f746
RG
387int vmapbuf (struct buf *);
388void vunmapbuf (struct buf *);
389void relpbuf (struct buf *, int *);
390void brelvp (struct buf *);
391void bgetvp (struct vnode *, struct buf *);
b153f746 392int allocbuf (struct buf *bp, int size);
41cbf46c 393int scan_all_buffers (int (*)(struct buf *, void *), void *);
1f1ea522 394void reassignbuf (struct buf *);
b153f746 395struct buf *trypbuf (int *);
984263bc
MD
396
397#endif /* _KERNEL */
398
399#endif /* !_SYS_BUF_H_ */